I don't think people have asked the right questions yet.

You say 25,000 accounts, but fail to mention how big their portfolios are? Or how frequently you will be repeating these calculations?

PDL is generally a way of speeding up complex math on very large numbers of similar calculations.

But 25,000 is not a large number. And averaging is not a complex calculation. By way of giving a feel, averaging 25k sets of 10 floats using straight Perl takes less 1/20th of a second:

[0] Perl> @a = map [ map{ rand( 1000 ) } 1 .. 10 ], 1 .. 25e3;;
[0] Perl> 
$t = time; 
push @$_, sum( @$_ ) / @$_ for @a; 
printf "Took %.6f\n", time() - $t;;
Took 0.045407
[download]

So, the questions you need to ask yourself are:

1. Do you need the speed?

   Is the need sufficient to warrant the extra complexity?

2. Do the actual calculations lend themselves to PDLs strengths?

   That's harder to answer, but from your description, I don't think they do.