I second the recommendation of PDL for meddling with numbers; on the statistics front I suggest R, I never used it with perl (I used it with python), but there are also perl-bindings. We were more than content, as it was easily capable of dealing with a very large data-set (30MB), SPSS couldn't even dream to work on (we had an SPSS license and tried it first, but it simply couldn't handle the amount of data per sample, let alone all samples...), STATS, which we considered switching to and had an evaluation license for, was of no use either; though I liked it a lot compared to SPSS...