diff --git a/Include/internal/pycore_interp_structs.h b/Include/internal/pycore_interp_structs.h
index badc97808c6132..9cdaa950e3479a 100644
--- a/Include/internal/pycore_interp_structs.h
+++ b/Include/internal/pycore_interp_structs.h
@@ -939,7 +939,7 @@ struct _is {
     struct _PyExecutorObject *executor_deletion_list_head;
     struct _PyExecutorObject *cold_executor;
     int executor_deletion_list_remaining_capacity;
-    size_t trace_run_counter;
+    size_t executor_creation_counter;
     _rare_events rare_events;
     PyDict_WatchCallback builtins_dict_watcher;
 
diff --git a/Include/internal/pycore_optimizer.h b/Include/internal/pycore_optimizer.h
index 685c39dcd65fb9..8ed5436eb6838c 100644
--- a/Include/internal/pycore_optimizer.h
+++ b/Include/internal/pycore_optimizer.h
@@ -90,8 +90,9 @@ PyAPI_FUNC(void) _Py_Executors_InvalidateCold(PyInterpreterState *interp);
 #endif
 
 // Used as the threshold to trigger executor invalidation when
-// trace_run_counter is greater than this value.
-#define JIT_CLEANUP_THRESHOLD 100000
+// executor_creation_counter is greater than this value.
+// This value is arbitrary and was not optimized.
+#define JIT_CLEANUP_THRESHOLD 1000
 
 #define TRACE_STACK_SIZE 5
 
diff --git a/Include/object.h b/Include/object.h
index 9585f4a1d67a52..7f4b35df3b6263 100644
--- a/Include/object.h
+++ b/Include/object.h
@@ -695,8 +695,13 @@ PyAPI_DATA(PyObject) _Py_NotImplementedStruct; /* Don't use this directly */
 #  define Py_NotImplemented (&_Py_NotImplementedStruct)
 #endif
 
-/* Macro for returning Py_NotImplemented from a function */
-#define Py_RETURN_NOTIMPLEMENTED return Py_NotImplemented
+/* Macro for returning Py_NotImplemented from a function. Only treat
+ * Py_NotImplemented as immortal in the limited C API 3.12 and newer. */
+#if defined(Py_LIMITED_API) && Py_LIMITED_API+0 < 0x030c0000
+#  define Py_RETURN_NOTIMPLEMENTED return Py_NewRef(Py_NotImplemented)
+#else
+#  define Py_RETURN_NOTIMPLEMENTED return Py_NotImplemented
+#endif
 
 /* Rich comparison opcodes */
 #define Py_LT 0
diff --git a/Lib/test/test_free_threading/test_bz2.py b/Lib/test/test_free_threading/test_bz2.py
new file mode 100644
index 00000000000000..0e09c64d5610a3
--- /dev/null
+++ b/Lib/test/test_free_threading/test_bz2.py
@@ -0,0 +1,53 @@
+import unittest
+
+from test.support import import_helper, threading_helper
+from test.support.threading_helper import run_concurrently
+
+bz2 = import_helper.import_module("bz2")
+from bz2 import BZ2Compressor, BZ2Decompressor
+
+from test.test_bz2 import ext_decompress, BaseTest
+
+
+NTHREADS = 10
+TEXT = BaseTest.TEXT
+
+
+@threading_helper.requires_working_threading()
+class TestBZ2(unittest.TestCase):
+    def test_compressor(self):
+        bz2c = BZ2Compressor()
+
+        def worker():
+            # it should return empty bytes as it buffers data internally
+            data = bz2c.compress(TEXT)
+            self.assertEqual(data, b"")
+
+        run_concurrently(worker_func=worker, nthreads=NTHREADS)
+        data = bz2c.flush()
+        # The decompressed data should be TEXT repeated NTHREADS times
+        decompressed = ext_decompress(data)
+        self.assertEqual(decompressed, TEXT * NTHREADS)
+
+    def test_decompressor(self):
+        chunk_size = 128
+        chunks = [bytes([ord("a") + i]) * chunk_size for i in range(NTHREADS)]
+        input_data = b"".join(chunks)
+        compressed = bz2.compress(input_data)
+
+        bz2d = BZ2Decompressor()
+        output = []
+
+        def worker():
+            data = bz2d.decompress(compressed, chunk_size)
+            self.assertEqual(len(data), chunk_size)
+            output.append(data)
+
+        run_concurrently(worker_func=worker, nthreads=NTHREADS)
+        self.assertEqual(len(output), NTHREADS)
+        # Verify the expected chunks (order doesn't matter due to append race)
+        self.assertEqual(set(output), set(chunks))
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/Lib/test/test_time.py b/Lib/test/test_time.py
index ebc25a589876a0..c7e81fff6f776b 100644
--- a/Lib/test/test_time.py
+++ b/Lib/test/test_time.py
@@ -759,7 +759,6 @@ class TestStrftime4dyear(_TestStrftimeYear, _Test4dYear, unittest.TestCase):
 
 class TestPytime(unittest.TestCase):
     @skip_if_buggy_ucrt_strfptime
-    @unittest.skipUnless(time._STRUCT_TM_ITEMS == 11, "needs tm_zone support")
     def test_localtime_timezone(self):
 
         # Get the localtime and examine it for the offset and zone.
@@ -794,14 +793,12 @@ def test_localtime_timezone(self):
         self.assertEqual(new_lt.tm_gmtoff, lt.tm_gmtoff)
         self.assertEqual(new_lt9.tm_zone, lt.tm_zone)
 
-    @unittest.skipUnless(time._STRUCT_TM_ITEMS == 11, "needs tm_zone support")
     def test_strptime_timezone(self):
         t = time.strptime("UTC", "%Z")
         self.assertEqual(t.tm_zone, 'UTC')
         t = time.strptime("+0500", "%z")
         self.assertEqual(t.tm_gmtoff, 5 * 3600)
 
-    @unittest.skipUnless(time._STRUCT_TM_ITEMS == 11, "needs tm_zone support")
     def test_short_times(self):
 
         import pickle
diff --git a/Misc/NEWS.d/next/C_API/2025-10-26-16-45-06.gh-issue-140487.fGOqss.rst b/Misc/NEWS.d/next/C_API/2025-10-26-16-45-06.gh-issue-140487.fGOqss.rst
new file mode 100644
index 00000000000000..16b0d9d4084ba0
--- /dev/null
+++ b/Misc/NEWS.d/next/C_API/2025-10-26-16-45-06.gh-issue-140487.fGOqss.rst
@@ -0,0 +1,2 @@
+Fix :c:macro:`Py_RETURN_NOTIMPLEMENTED` in limited C API 3.11 and older:
+don't treat ``Py_NotImplemented`` as immortal. Patch by Victor Stinner.
diff --git a/Modules/_bz2module.c b/Modules/_bz2module.c
index 9721b493a19956..452b88dfed29ce 100644
--- a/Modules/_bz2module.c
+++ b/Modules/_bz2module.c
@@ -97,20 +97,11 @@ OutputBuffer_OnError(_BlocksOutputBuffer *buffer)
 #endif  /* ! BZ_CONFIG_ERROR */
 
 
-#define ACQUIRE_LOCK(obj) do { \
-    if (!PyThread_acquire_lock((obj)->lock, 0)) { \
-        Py_BEGIN_ALLOW_THREADS \
-        PyThread_acquire_lock((obj)->lock, 1); \
-        Py_END_ALLOW_THREADS \
-    } } while (0)
-#define RELEASE_LOCK(obj) PyThread_release_lock((obj)->lock)
-
-
 typedef struct {
     PyObject_HEAD
     bz_stream bzs;
     int flushed;
-    PyThread_type_lock lock;
+    PyMutex mutex;
 } BZ2Compressor;
 
 typedef struct {
@@ -126,7 +117,7 @@ typedef struct {
        separately. Conversion and looping is encapsulated in
        decompress_buf() */
     size_t bzs_avail_in_real;
-    PyThread_type_lock lock;
+    PyMutex mutex;
 } BZ2Decompressor;
 
 #define _BZ2Compressor_CAST(op)     ((BZ2Compressor *)(op))
@@ -271,12 +262,12 @@ _bz2_BZ2Compressor_compress_impl(BZ2Compressor *self, Py_buffer *data)
 {
     PyObject *result = NULL;
 
-    ACQUIRE_LOCK(self);
+    PyMutex_Lock(&self->mutex);
     if (self->flushed)
         PyErr_SetString(PyExc_ValueError, "Compressor has been flushed");
     else
         result = compress(self, data->buf, data->len, BZ_RUN);
-    RELEASE_LOCK(self);
+    PyMutex_Unlock(&self->mutex);
     return result;
 }
 
@@ -296,14 +287,14 @@ _bz2_BZ2Compressor_flush_impl(BZ2Compressor *self)
 {
     PyObject *result = NULL;
 
-    ACQUIRE_LOCK(self);
+    PyMutex_Lock(&self->mutex);
     if (self->flushed)
         PyErr_SetString(PyExc_ValueError, "Repeated call to flush()");
     else {
         self->flushed = 1;
         result = compress(self, NULL, 0, BZ_FINISH);
     }
-    RELEASE_LOCK(self);
+    PyMutex_Unlock(&self->mutex);
     return result;
 }
 
@@ -357,13 +348,7 @@ _bz2_BZ2Compressor_impl(PyTypeObject *type, int compresslevel)
         return NULL;
     }
 
-    self->lock = PyThread_allocate_lock();
-    if (self->lock == NULL) {
-        Py_DECREF(self);
-        PyErr_SetString(PyExc_MemoryError, "Unable to allocate lock");
-        return NULL;
-    }
-
+    self->mutex = (PyMutex){0};
     self->bzs.opaque = NULL;
     self->bzs.bzalloc = BZ2_Malloc;
     self->bzs.bzfree = BZ2_Free;
@@ -382,10 +367,8 @@ static void
 BZ2Compressor_dealloc(PyObject *op)
 {
     BZ2Compressor *self = _BZ2Compressor_CAST(op);
+    assert(!PyMutex_IsLocked(&self->mutex));
     BZ2_bzCompressEnd(&self->bzs);
-    if (self->lock != NULL) {
-        PyThread_free_lock(self->lock);
-    }
     PyTypeObject *tp = Py_TYPE(self);
     tp->tp_free((PyObject *)self);
     Py_DECREF(tp);
@@ -619,12 +602,12 @@ _bz2_BZ2Decompressor_decompress_impl(BZ2Decompressor *self, Py_buffer *data,
 {
     PyObject *result = NULL;
 
-    ACQUIRE_LOCK(self);
+    PyMutex_Lock(&self->mutex);
     if (self->eof)
         PyErr_SetString(PyExc_EOFError, "End of stream already reached");
     else
         result = decompress(self, data->buf, data->len, max_length);
-    RELEASE_LOCK(self);
+    PyMutex_Unlock(&self->mutex);
     return result;
 }
 
@@ -650,13 +633,7 @@ _bz2_BZ2Decompressor_impl(PyTypeObject *type)
         return NULL;
     }
 
-    self->lock = PyThread_allocate_lock();
-    if (self->lock == NULL) {
-        Py_DECREF(self);
-        PyErr_SetString(PyExc_MemoryError, "Unable to allocate lock");
-        return NULL;
-    }
-
+    self->mutex = (PyMutex){0};
     self->needs_input = 1;
     self->bzs_avail_in_real = 0;
     self->input_buffer = NULL;
@@ -678,15 +655,13 @@ static void
 BZ2Decompressor_dealloc(PyObject *op)
 {
     BZ2Decompressor *self = _BZ2Decompressor_CAST(op);
+    assert(!PyMutex_IsLocked(&self->mutex));
 
     if(self->input_buffer != NULL) {
         PyMem_Free(self->input_buffer);
     }
     BZ2_bzDecompressEnd(&self->bzs);
     Py_CLEAR(self->unused_data);
-    if (self->lock != NULL) {
-        PyThread_free_lock(self->lock);
-    }
 
     PyTypeObject *tp = Py_TYPE(self);
     tp->tp_free((PyObject *)self);
diff --git a/Python/bytecodes.c b/Python/bytecodes.c
index f9f14322df0a5e..6ebd9ebdfce1bb 100644
--- a/Python/bytecodes.c
+++ b/Python/bytecodes.c
@@ -11,7 +11,6 @@
 #include "pycore_audit.h"         // _PySys_Audit()
 #include "pycore_backoff.h"
 #include "pycore_cell.h"          // PyCell_GetRef()
-#include "pycore_ceval.h"
 #include "pycore_code.h"
 #include "pycore_emscripten_signal.h"  // _Py_CHECK_EMSCRIPTEN_SIGNALS
 #include "pycore_function.h"
@@ -5362,10 +5361,6 @@ dummy_func(
 
         tier2 op(_MAKE_WARM, (--)) {
             current_executor->vm_data.warm = true;
-            // It's okay if this ends up going negative.
-            if (--tstate->interp->trace_run_counter == 0) {
-                _Py_set_eval_breaker_bit(tstate, _PY_EVAL_JIT_INVALIDATE_COLD_BIT);
-            }
         }
 
         tier2 op(_FATAL_ERROR, (--)) {
diff --git a/Python/ceval_gil.c b/Python/ceval_gil.c
index 6bf64868cbb2d3..9b6506ac3326b3 100644
--- a/Python/ceval_gil.c
+++ b/Python/ceval_gil.c
@@ -1398,7 +1398,7 @@ _Py_HandlePending(PyThreadState *tstate)
     if ((breaker & _PY_EVAL_JIT_INVALIDATE_COLD_BIT) != 0) {
         _Py_unset_eval_breaker_bit(tstate, _PY_EVAL_JIT_INVALIDATE_COLD_BIT);
         _Py_Executors_InvalidateCold(tstate->interp);
-        tstate->interp->trace_run_counter = JIT_CLEANUP_THRESHOLD;
+        tstate->interp->executor_creation_counter = JIT_CLEANUP_THRESHOLD;
     }
 
     /* GIL drop request */
diff --git a/Python/executor_cases.c.h b/Python/executor_cases.c.h
index 0e4d86463761a0..9ce0a9f8a4d87b 100644
--- a/Python/executor_cases.c.h
+++ b/Python/executor_cases.c.h
@@ -7409,9 +7409,6 @@
 
         case _MAKE_WARM: {
             current_executor->vm_data.warm = true;
-            if (--tstate->interp->trace_run_counter == 0) {
-                _Py_set_eval_breaker_bit(tstate, _PY_EVAL_JIT_INVALIDATE_COLD_BIT);
-            }
             break;
         }
 
diff --git a/Python/import.c b/Python/import.c
index d01c4d478283ff..45206b46793846 100644
--- a/Python/import.c
+++ b/Python/import.c
@@ -2358,6 +2358,7 @@ create_builtin(PyThreadState *tstate, PyObject *name, PyObject *spec)
     for (struct _inittab *p = INITTAB; p->name != NULL; p++) {
         if (_PyUnicode_EqualToASCIIString(info.name, p->name)) {
             found = p;
+            break;
         }
     }
     if (found == NULL) {
diff --git a/Python/optimizer.c b/Python/optimizer.c
index 6ad9124744859a..f44f8a9614b846 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -6,6 +6,7 @@
 #include "pycore_interp.h"
 #include "pycore_backoff.h"
 #include "pycore_bitutils.h"        // _Py_popcount32()
+#include "pycore_ceval.h"       // _Py_set_eval_breaker_bit
 #include "pycore_code.h"            // _Py_GetBaseCodeUnit
 #include "pycore_function.h"        // _PyFunction_LookupByVersion()
 #include "pycore_interpframe.h"
@@ -1343,6 +1344,14 @@ uop_optimize(
         return -1;
     }
     assert(length <= UOP_MAX_TRACE_LENGTH);
+
+    // Check executor coldness
+    PyThreadState *tstate = PyThreadState_Get();
+    // It's okay if this ends up going negative.
+    if (--tstate->interp->executor_creation_counter == 0) {
+        _Py_set_eval_breaker_bit(tstate, _PY_EVAL_JIT_INVALIDATE_COLD_BIT);
+    }
+
     *exec_ptr = executor;
     return 1;
 }
diff --git a/Python/pystate.c b/Python/pystate.c
index 5d0927c6c08196..2141e842a37d2f 100644
--- a/Python/pystate.c
+++ b/Python/pystate.c
@@ -571,7 +571,7 @@ init_interpreter(PyInterpreterState *interp,
     interp->executor_list_head = NULL;
     interp->executor_deletion_list_head = NULL;
     interp->executor_deletion_list_remaining_capacity = 0;
-    interp->trace_run_counter = JIT_CLEANUP_THRESHOLD;
+    interp->executor_creation_counter = JIT_CLEANUP_THRESHOLD;
     if (interp != &runtime->_main_interpreter) {
         /* Fix the self-referential, statically initialized fields. */
         interp->dtoa = (struct _dtoa_state)_dtoa_state_INIT(interp);