It may not be the fastest, but something like Common Lisp has extensive numeric capabilities built-in like computing with floats, bignums, complex, ratios, ... Extensive mathematical software has been written in Lisp like Reduce (written in Standard Lisp), Macsyma, Axiom, ... In education for a while something like Derive, MuSimp/MuMath, ... was used. Derive should also have been used in pocket calculators, which would be in the spirit of Lisperati1000.
If you stick to floats and arrays of floats, Fortran is probably still faster.
But Common Lisp and fully conformant Schemes have an extensive numeric tower including arbitrary precision integers, rationals, and complex numbers built in, making Lisp useful for some kinds of numeric computing that would be cumbersome even in Fortran.
Plus, I once heard of a guy who wrote an FFT implementation in Gambit Scheme that beat FFTW in speed...
Lisp was the favored language for programming AI back in the day. Though the kind of AI problems focused on then was much more symbolic-themed than numerical themed as they are now. Lisp is highly regarded when solving complex problems, though would probably not get hailed as the fastest language.
> But if you need some complex algorithms — particularly algorithms that do a lot of heavy mathematical lifting — then Lisp is the ideal choice.
Is this right? I never thought of LISP as good fit for numerical processing.