I did a fair bit of similar work at Stack Overflow with @codinghorror.
I like the idea, and in high scale communities it can really help tidying stuff up.
Thing is, to build this kind of feature properly you need to have a corpus of data and problem patterns you are trying to eradicate. Otherwise we would be building this based on guess work that very much risks an unbalanced approach.
Keep in mind we already handle some of the more trivial ones with our entropy checks, however its a system we should be constantly refining based on real-world data.