[trishume]

Idea: I wonder if it would be profitable to have a mode for Pyston/PyPy/ZiPy/Whatever that disables threads, or perhaps an annotation that stops other threads from being scheduled inside certain loops.

This would allow so many checks to be hoisted out of the loop like LuaJIT does, that otherwise can't be hoisted because in theory some thread could re-assign os.path in the middle of it. If this could provide a 2-5x speedup, in most scenarios it would outweigh the performance benefit of parallelism, as well as being easier than paralellizing.