Does anyone have a suggestion on how many runs should you do to get a good representative sample of the performance. My understanding is that because of varying network conditions, its important to have a larger number. Somewhere I recall reading that something like 100 runs would increase the confidence level.
Also, WebPagetest uses the median rather than the average. As I understand it, a median is better when the data is skewed. Is this the reason why for web performance using the median is the better measurement?