I've just been running the universe test on one of my machines and I've noticed a bug in the parser which is similar to the scale problem identified in the md-gromacs test-set.
On fast machines, the result given isn't in seconds anymore, it's in milliseconds.
This leads to some stupid results for fast machines like the one I tested on. It reported times of 7000ms which the parser interprets as 7000s.
I'm looking into the parser now to work out how best to fix it.