Provides some clearly measurable speedup for the modified test.
E.g. 1p_richards1p_1d3d, the solver time halves with four proceses.
Not a good as scalability of assembly but this is expected.