For profilng IO-bound Python code, I'm finding that cProfile produces no measurable slowdown. Impressive.