High Performance Python Tutorial by Ian Ozsvald
(hint: look for the PDF file or you'll miss the wealth of it — all 46 pages!)
Normally I'm cautious about performance metrics — they have a tendency to compare dissimilar technologies and often avoid important details (like standard deviations) — but with this article, by Ian Ozsvald, I am quite thrilled with the completeness of his results.
He goes through a number of different scenarios implementing an algorithm in python and the available ways to tune the code using traditional methods (numpy and pypy) and non-traditional methods (pyCUDA, cython, shedskin, and ParallelPython), all the while comparing the relative performance and difficulty of implementing the code.
I wouldn't mind seeing more articles like this!