> I'd like to see some weighted reputation system used instead, where academics can give their "vote" to papers they like, giving it a total weight, however such systems have been proposed in the past and haven't got much traction.

A reddit-clone with strong verification/validation of user accounts seems like it would fit the bill. I can imagine a system with a small monthly fee to participate, subreddits with paid moderators (read: reviewers), voting system that would float good papers to the top, and of course comments. The pay-to-participate barrier would hopefully keep the quality high but at a minimum cover the hosting costs. And run by a non-profit foundation.