For example, we caught a bunch of users who were gaming it by deleting any comments that brought their average down. We fixed that by treating deleted comments as comments of score 0 for average-computing purposes. It's an example of how, once you publish a metric, people start to care about it and do things based on it, regardless of how meaningful it is.
Comment average used to be used by a few algorithms (like comment ranking) but we turned that off as an experiment a while ago and nothing seemed to get worse. If anything, I think it may have helped a little.