Wow, I've never seen that before. I'm also not sure I'd say a solution that uses bytecode rewriting is really working within the language, but that's a pretty interesting package anyhow.
Well, normally I would agree - but if its entirely abstracted away in a library (and the library is well written) then it might sort-of-kind-of-maybe be ok.
Python itself does varying degrees of bytecode optimisation depending on which version you're using.