Interning the __module__ string causes small performance hit:

$ ./python -m timeit -s "from itertools import chain" -- "chain.__module__; chain.__module__; chain.__module__; chain.__module__; chain.__module__; chain.__module__; chain.__module__; chain.__module__; chain.__module__; chain.__module__"

Unpatched:  1.93 usec per loop
Patched:    4.09 usec per loop

This can be avoided if cache created string in type's __dict__. Following patch makes __module__ retrieving for non-heap types as fast as for heap types:

Patched2:   0.871 usec per loop