Thanks for writing this article. I have been using the benchmark library for a few weeks now and wondering how best to interprete the "before" and "after" results. I did not think to statistically analyse my iterations- this article has given me lots to consider.