It appears he throws out the highest rated comment when computing the average. I assume that's because of what you're getting at: the distribution of comment scores tend to look like a power law, so the arithmetic mean of all the scores isn't really the right number to use.