(you will notice the numbers are different on the HN leaders page for people, I think perhaps searchyc is an all-time average?)
Like the frontpage ranking algorithm, the comment ranking algorithm also considers recency.
An alternative way to think about it is that it adds more stability to the system, for better or for worse.
The whole thing has a circular dependency: if ranking acts as a filter, then higher rank means more readers, and that in turn means more upvotes and so a higher ranking. It is probably not so much the top being undeserving, but that deserving stuff gets missed.
This seems to be a fundamental weakness in all similar 'ranking' systems, but I am not sure of the full character and implications...
(There is certainly a substantial component of being an automated system of 'social-proof'. And the filtering can never be entirely effective: if you show everyone only the good stuff, the filtering would not get done at all.)
Otherwise they get poor real fast.
So someone with a huge number of highly rated comments, could make a lot of dumb comments before it started hurting them.
I'm sure pg will come up with an appropriate formula.
Does this create a rich-get-richer type scenario, where users with high averages get more exposure, thus leading to higher averages?