A system like this would still not account for outliers, nor deal well with temporal changes. People can be completely well behaved for long periods of time and suddenly really hit rock bottom. If you're the person who they are interacting with at the time, the trust system fails. Sure they get a lowered rating now, but then you have to ask the system, was this person who had a bad interaction being honest and how much of a ding is this to a long record of good ratings.

There's no catch all, even a single, centralized system wouldn't be very successful at dealing with the outliers in human behavior on top of the questionable honesty of the people adding ratings to the system in the first place.