This is another example of how a raw arithmetic mean can be highly deceptive. Looking at the median performance, or the mean +/- standard deviation, is much more informative.

A good example (which I picked up from Gnuplot in Action) is this: you're organising a marathon and need to know when to have end-of-race refreshments ready. You load end time records into a database, then take an average (say ... 1 hour, 30 minutes). You plan to put out drinks at 1 hour 15 minutes, expecting the bulk of the runners to arrive then.

The big day arrives. Suddenly, around the 50 minute mark, runners start arriving in surprisingly large quantities. It's a disaster, there's nothing for them to drink, and afterwards you live a life of chagrin and go to your grave humiliated (marathon runners are pretty serious people).

What the mean obscured was that the marathon times were bimodally distributed. About 20% of runners, serious marathoners, arrived "early", and the rest of the runners much later. If you'd thrown it up on a graph or looked at some other figures, you'd have known. But you were deceived by the easiness of the mean.