As someone who frequents many websites with some variation of a rating system, I thought I'd chime in from the user perspective.

The type of "best of" lists that I like are the ones that give me not just a ranked list (ie: the top 10 whatsits) but the ratings and some indication of the number of votes. That helps me, as a user, to decide how accurate the rating is.

I'm a fan of the practice of displaying the rating in a different way at different confidence levels (ie: bigger, bolder, different colors) to convey the data. Think: 'tag cloud'.

You list "Offloads the real sorting and determining of value onto the user" as a problem with the first scheme. I can see that it could be if you go with a standard deviation as the indicator of the number of votes. Most people wouldn't have the first idea of what to do with that information.

I think the average vote, along with the number of votes (or some graphical indication thereof) and an option to see the breakdown of votes is a good system. That way the users can see, in the case of an item with a small number of votes, if a single vote has skewed the average.

That's enough blathering on from me....I hope it at least gave you some ideas. Please post again and let us all know what you decide on in the end. I'm very curious!

Cheers!