I guess the way to benchmark browsing to get most realistic numbers would be to record normal browser behavior of a decently-sized sample of people (say 1000) and automatically repeat the behavior of a representative subsample (10-50 may be enough).

This would take forever and cost quite a lot of money, however, so the much more reasonable solution is to simply turn off the cache for everything.

I don't agree with Apple or Techcrunch that CR's methodology is wrong -- pretty much anyone would make that same decision because the alternative is too much work for too little reward.