As dagw mentions - treat it as a feature. I taught High Performance Python at PyCon, I've published all the src and a 55 page PDF covering a range of techniques to optimise a Mandelbrot example (including cython, shedskin, numpy, multiprocessing, pycuda): http://ianozsvald.com/2012/03/18/high-performance-python-1-f... With some languages this is painful, with Python you can easily profile (cProfile, line_profiler) and then work on easy or complex optimisations (both within Python and using external tools) - great flexibility for when you really need it.