For long reads, directly or segment using SPOA?

For some long reads (10kbp), if their left ends are nearly aligned, should we directly use SPOA to generate a consensus sequence, or adopt a segmented SPOA approach (e.g., segment the reads into 200bp windows first and then merge the results via SPOA)?