Shed Skin - An Optimizing Python-to-C++ Compiler
shed-skin.blogspot.com
shed-skin.blogspot.com
I guess that shows how well tuned Python's sets/dicts are -- gprof suggested most time was spent in Shedskin's sets, looking up keys/comparing keys; even using -O3 did not help.
I currently use Pyrex (Cython is a newer version of it), successfully to write 5% of the code in its Pythony C. It's rare in my case that I can find a spot which can be easily rewritten in C; it may be 5% of the code that's time-intensive but it's 5% using all sorts of complex data structures created by the other 95%.
I don't think comparing to Unladen or other JIT tech is appropriate. The real use for ss (in my opinion) is to replace the prototype-in-python, optimize-in-C++ workflow. Now it's prototype-in-python, optimize-in-python. Yes a few edge cases need handling (esp if you want the real speedup; avoid things like lists with random types) but the end result should be greatly improved productivity and code readability.