This problem seems to be specific to OS-X. I haven't reproduced it on a Linux cluster. Monitoring PIDs, I can see that the Python threads are constant throughout the run, but new threads are constantly being spawned by rcolgem. Eventually this locks up the run.