In longer tests with g=2, DH exchange does not become quite as fast

as expected -- maybe it's the different processor, maybe my
previous timings were too inaccurate.