What the fastest algorithm will be depends on a lot of things, for instance whether you will have more failures than successes. Size of the words are important too.

Here's one algorithm that benefits from "early failures", as soon as a letter in the second word is processed for which there's no corresponding letter in the first word, it returns. It's also using an array to count, not a hash.

sub match {
    my @h;
    ++ $h [ord]                for split // => shift;
    -- $h [ord] < 0 and return for split // => shift;
    1;
}
[download]

It's a small (and untested) subroutine. The benefit of using an array, and splitting on an empty regex is that it can be ported to C trivially. If speed is really important, I'd consider using Inline::C.

Note also that you could archieve major speedup by not calling the match function a million times - instead call it with a list of pairs of arguments to process. Calling a function is not a cheap task. However, you need to be careful and not kill the gained performance by using too much memory.

Abigail