Also the second run probably used stuff that was put in the npm cache during the first run. Not sure how good a test this is.

A good read: "Programmers need to learn statistics or I will kill them all" http://zedshaw.com/archive/programmers-need-to-learn-statist...