I agree that average karma is a very poor metric, and I found that my usage of HN changed drastically when pg started showing it more prominently. One optimizes for average karma by not commenting on stories that are older or less popular, as those are likely to produce a 1 or 2 rating due to lack of eyeballs. However such comments are often interesting and useful contributions, and it hurts HN to discourage them.
A possible solution is to scale by the number of people who look at the comment, although this might be difficult to do well. You could probably get better results by estimating a regression containing the following variables: age of post, score of post, number of comments, average score of comments in the thread, and the depth of the tree in which the comment was posted to get a good determination of how many points an average comment in that situation would get.