Numpy support is improving but incomplete, see http://morepypy.blogspot.com/2015/02/linalg-support-in-pypyn... - and the performance is not going to be any better than stock numpy with CPython. If your code spends all its time in C extensions, PyPy doesn't buy you anything since its API compatibility layer is slow and incomplete. You should probably look into Julia.
I mainly use Scikit Learn, theano, numpy and pandas; is PyPy able to work with the above, and likely to give any speedups at this stage?