FFI into lean C isn't some perf panacea either, beyond the overhead you're also depriving yourself of interprocedural optimization and other Good Things from the native space.