Cool. This is a naïve question, but how come "gmp does better than Gambit-C on bigint multiply, using FFTs."? (This has the potential to derail this conversation, feel free to break off.)