So what you're really doing is (in parallel) finding the decoherence points... These branching/decision points are much more easily found by instrumenting the code as in AFL.

No need to parallelise at all, the compiler gives you that information.

What you have seems like a neat trick, but not an efficient way of actually fuzzing.