The FFI reasonings are OK, I understand them. I didn't think in those terms originally. (Thought I asked about writing a C app not a C library...)
But speed and C are othogonal nowadays (at least if productivity is in the picture). It is so much simpler to write correct programs leveraging multicore environments an/or asynchronity in other environments, that it is not productive to chose C for speed. (In terms of development costs. Ofc. there are special cases when it might be worth, I'm talking about the general situation)