Because most of the really huge scientific Python libraries are written as wrappers over lower-level language code, I'd be curious to what extent speeding up Python by, say, 10% would speed up "normal" scientific Python code on average. 1%? 5%?