For unicode at the very least it's not an improvement at all. With the patch mentioned that does encode it's also not an improvement at all. Prove as in reason about the function in C and make sure it does not do any conditionals depending on the input data. This is much easier than it is in Python.

We did this exercise for PyPy once, just for the sake of it. We looked at generated IR and made sure a comparison is not leaking any data.

As far as the function goes right now - I don't know. For now following the entire code of long_bitwise is a lot of effort - I genuinely can't say that it'll be the same for all numbers of 0-255. Can you? It's easier with low-level language simply (And yes, this is one of the few cases where I would argue it makes sense to implement something in C :)