Interesting - will check out Suarez. You're right that doing this in a privacy protecting manner is going to be Hard: by definition it's asking users to disclose their views - although this doesn't have to be to the general public; it could be an opt in thing: eg "Alice wants to subscribe to your reputation data - Yes/No". Storing the reputation data so that it doesn't become a honeypot of metadata is harder however.
There is absolutely the problem that widely trusted people would use their position to push an agenda, deliberately or unintentionally. This is inevitable, as far as I can see, given we are assuming there are no moral absolutes here. So the best we can do is to ensure that users have visibility on the amount and shape of content being filtered out, access to whichever 'authorities' they chose to trust, and make it easy to view the world from different viewpoints.
Fun problems! I can't wait for Matrix to properly focus on it.
Two principal problems are probably Big Money gaming the system and people not wanting to publicly disclose their views. In the worst case your scheme might end up serving age-old staples of propaganda: Falsely implying a consensus and the establishment of authority (of the most trusted users).
I'll keep a look out for how you try to deal with that in the future.