Amazon might not be the best example - loading http://cssstats.com/stats?url=http%3A%2F%2Famazon.com#specif... multiple times has generated multiple, very different graphs. I wonder if the CSSStats bot has been thrown into more than one A/B experiments.