But strangely they don't - PyPy has hardly gained traction (albeit the python2->python3 switch didn't help) and looking at the benchmarks that's more like 5-7x performance.
I suspect that most often, in places where performance matters enough in a way that would warranted refactoring a code-base from cpython to PyPy, they already rewrote the botte-neck parts in C anyways.
The issue I've usually had with it is just small things – maybe an module we depend on doesn't work with PyPy yet, maybe PyPy3 isn't 3.3-compatible yet and can't use "yield from", which we use in our code or which our libraries use.
Dealing with those sort of small issues (and/or waiting for PyPy to fix them), worrying about whether the project I'm currently working on can use PyPy or if I need to use CPython instead, etc makes me stop worrying about trying to use it after a while. In my dev environment at least.
I do really love what the PyPy guys are doing though, and when it works it works damn well. Building it from source is also super pretty.