The bignum performance in Gambit-C 4.0beta is about half that of gmp-4.1.3---if gmp takes time 1, then Gambit generally takes time 2. (GCD, and hence rational arithmetic, in gambit is slower than that at the moment, but ...) If I substituted a different FFT routine in C for the current FFT routine in Scheme, then much of the difference would be made up---if gmp takes time 1, gambit would take about time 4/3. This is for integers of size >= about 30,000 bits.
Is this if interest to anyone? Is it worth the hack?
Brad