I’ve been meaning to do it for a while but just got around to changing the data in the summary results table from averages to medians (a lot more useful if you have outliers).
This only impacts tests with multiple runs and the averages are still available through the API if you were using those.
Here is an example: http://www.webpagetest.org/result/101029_9Z0V/
Instead of a single run identified as being “closest to the average” it independently identifies the median results for first and repeat views.
In the case of an even number of test runs the faster of the two runs in the middle is picked.