I probably should have looked at quartiles, since the distribution has a lower limit but not an upper limit, and even a few bad runs can drive the sample mean up crazy high