I have not used this, and I do not know about the raw bandwidth issue, but you may like to try GSL (Gnu Scientific Library) with ATLAS for enhanced BLAS matrix operations. See "Alternative BLAS Libraries" on the GSL page. Perhaps people who do more matrix multiplication can compare this to the other two packages mentioned above.

However the GSL page mentions PDL::GSL which I can't seem to find. And I don't know if Math::GSL covers matrix multiplication (you'd think it would..). On the other hand, if the important part is the BLAS implementation and you are using ATLAS, why not just use ATLAS either with Inline::C or better yet from command line, skipping all the to-from perl overhead.

From http://math-atlas.sourceforge.net/,
The ATLAS (Automatically Tuned Linear Algebra Software) project is an ongoing research effort focusing on applying empirical techniques in order to provide portable performance. At present, it provides C and Fortran77 interfaces to a portably efficient BLAS implementation, as well as a few routines from LAPACK.

If someone else could comment on this I would appreciate it. Very likely the PDL people are well aware of ATLAS too.