I remember that Louis Helm (one of the creators of SoB  which uses the same algorithmic implementation as Prime95) once said that an implementation in Java would be roughly 30 times slower.
I think the reason here is similar: The speed difference (incidently 30 times as well) supposely comes from the fact that Mathematica's routines are not handoptimized for this exact purpose and architecture (does Mathematica use SSE2/FFTs/IBDWT?). In addition, I believe some "tricks" prime95 uses can only be applied due to special properties of mersenne numbers. Mathematica can't take that into account, I guess.
Last fiddled with by Mystwalker on 20050210 at 14:52
