I was interested in what the uncertainties were in benchmarks taken in Nexuiz. I made some calculations and have put together an OpenDocument spreadsheet (made in OOo, should work in others, if not give the appropriate projects bugzilla a nudge) for calculating these uncertainties:
http://www.rofltech.co.uk/media/nexuiz_ ... inties.ods

Run a benchmark then copy and paste the necessary values into the spreadsheet to get uncertainties in the min/avg/max values. Min and Max aren't comparable to Avg at all as they are only for that individual frame, not any frame that may have been min/max.

These are useful figures if you're interested in seeing how much variation there is in results. If you play with lots of settings then any setting that gives a performance change less that the uncertainty is not useful as it is not statistically significant.

If you would like to understand what any of this actually means, take a Physics degree.

Edit: This only works on Nexuiz before 2.3. In 2.3 the measurements changed and are incomparable.