Re: the original convo, they're stuck on the fact that arithmetic mean isn't terribly valid on a skewed distribution (and response/completion distributions are generally positive-skewed because of the firm/hard lower bound on task completion).
However, I do get what you're trying to do re: identify whether the problem is related to outliers. But in general, comparing to median has a lot more validity. Median isn't as sensitive to skew, and will be closer to the peak. For your purposes it probably wouldn't be a lot different but you wouldn't have pushed the "it's wrong" button.
All said, though, even comparing 50th (median) to 99th is pretty coarse. I'd probably be looking somewhere closer to 75th percentile for a comparison. Basically, you'd want to guess what percent might reasonably be affected by performance spikes and compare from there.
However, I do get what you're trying to do re: identify whether the problem is related to outliers. But in general, comparing to median has a lot more validity. Median isn't as sensitive to skew, and will be closer to the peak. For your purposes it probably wouldn't be a lot different but you wouldn't have pushed the "it's wrong" button.
All said, though, even comparing 50th (median) to 99th is pretty coarse. I'd probably be looking somewhere closer to 75th percentile for a comparison. Basically, you'd want to guess what percent might reasonably be affected by performance spikes and compare from there.