(Importing from bugs-everywhere)

Can criterion help us to figure out if our benchmarks have any kind of 
validity from a statistical point of view?

Comments from the bug-tracker

This could even lead to criterion/progression integration where we feed
this cached-on-disk data to the libraries.

The idea is to accumulate enough data over the long run (say over several
months) to do statistically interesting things.  Every time you run dars-
benchmark,
we squirrel away some new data in our archives and we get a better idea 
on the
variance, etc.