Numpy not using optimized multiplication routines

I'm trying to use python-numpy on 9.04 and simple benchmark computations are taking an extremely long time, I.E. a random 5000x5000 array takes over 3 minutes to multiply by itself when using dot(). Matlab equivalents take ~10-20s.

I suspect it's because the package isn't using ATLAS/BLAS/etc. for the computation, but I'm not entirely sure how to verify/fix it, short of building and installing numpy and it's dependencies myself.

Both blas and lapack packages are installed (atlas is not):

$ dpkg --get-selections | grep blas
libblas3gf install
$ dpkg --get-selections | grep lapack
liblapack3gf install