Expected: two successful benchmark groups, five timed runs each. Means depend on machine load; slow should normally exceed fast.
