As for breaking apart reviews into categories that also has its problem. Zagat used to rate restaurants in 5 categories and it ended up requiring people to lie if they wanted to convey the message "do not go here".
Sophisticated scoring might also bring common critiques, like ratings being biased or opaque (or just confusing). I recall Netflix had a issue when they made their star rating a prediction of what you would rate it (based on your previous viewing habits). It confused a lot of people who thought it was simply the average star rating. They since switched to percent-match as a result, and dropped star-reviews in favor of thumbs-up/thumbs-down.
Come to think of it, your suggestion may lead to something like percent-match for apps...though I suspect rating apps along dimensions the way Netflix does TV/movies may be tricky. Maybe different dimensions for different categories in the app store?
I'm still intrigued by a distributed approach that filters opinion through the sieve of the social network...I think it has more promise than centralized rating systems that attract abuse.
Federated/distributed systems where the network is itself the filter seem more promising to me, even though they tend to come at the expense of usability.