zlacker

The Kuro5hin solution of displaying averages plus number of ratings might give information in that direction, factoring out into two different numbers the average quality people thought the comment had, from the number of people who rated it. Not quite the same as a threshhold maximum, but it produces an asymptotic maximum instead. It's also, imo, useful to have them as two separate numbers, because they convey semantically different information.

>>_delir+(OP)
I like this idea. As another commenter pointed out, we seek out external validation (whether we like it or not). However, there's good and bad kinds of external validation: looking for popularity (total number of upvotes) vs. more direct feedback from people you respect (Kuro5hin model: a score with a name attached to it) works better. The fact that a comment would be rated rather than voted on Kuro5hin (with a maximum rating) also has the capping effect.

>>_delir+(OP)
Trimean or winsorized mean could be useful in this case.