adjusting mpi tests

I'm looking for what gives the systemic errors in the contact tests between the
sequential and the parallel implementations. Seems that size parity is
important, and that errors only occur in the highest frequency components.