At some point, the divide and conquer approach has so much overhead that the simple solutions are better. fork() is especially expensive, as once you get a few processes out there, context switching between them will take more time than plodding along in a for loop (which will probably optimize to a handful of machine code expressions anyway).

I really like the xor solution someone else pointed out, and you could probably do the 'x characters at a time' on it, too.