I have no idea. Assuming you go with the CFFI route (reusing our C API), you'll still have to write a lot of conversion code in Python and deal with generate messages (not sure if the generated Python messages will work as-is). It might be a considerable effort, but I don't know for sure.
W.r.t. parallel thread execution: PyPy suffers from GIL (like e.g. CPython does) and the GIL-free pypy-stm is highly experimental.