Thought experiment: how does one quantify the 'best' posts on a discussion topic? Discourse has some ideas about this, and I bet you do too.
"Best" posts--those which are a good source of information--have a high view-to-interaction ratio. An inflammatory post on a forum or Reddit tends to see more replies than the average. A good reference piece or poignant story seems more likely to stand on its own--it will have more views than most posts, and have a high view/reply ratio.
What separates a good post from an insignificant one is baseline views. If a thread has been opened 10000 times and everyone has read a certain comment, it's likely to be informative or controversial.
The "best of", even within a single topic/thread, is also per-user and temporal. The best-of in a thread to me is no longer a best-of when I've fully internalized it. I really value Reddit's collapse button/link for this reason.
To answer your question, I'd have to agree with what was already written by eric-hu, except to say that bad or inflammatory posts also create a lot of response (trolls, as it were). I know it's maybe tired or clunky, but I think a forum can have both entertaining and useful replies, and thus should have a voting mechanism to indicate which are which. Yes, interaction can play a role in how important they are, and to maybe help massage ranking, but I think people are happy with and familiar with a system of voting for content.
That said, a token "good post" should be able to stand for itself. I envision a system where the top posts (be they replies to other posts or posts in and of themselves) sit flat among each other, ranked by their usefulness or entertainment value.
I think a forum moderator's real value in this situation is not to mess with the community's decisiveness, but to resolve matters of content relevance (to keep a community focused).