My understanding is that the performance issues it has are more big-O issues of the problem space than issues that would meaningfully be resolved by the multiplicative speedup of moving to a more efficient language (the point at which they start to bite would shift a bit, though.)