Didn't Google release a study recently that said SMART was pretty much ineffective?
Also see this bit from Bill's notes: "Overall, from his [Luiz'] experience, he feels that SMART is inadequate to predict drive failures. It might be useful, though, in predicting population-wide trends."
I was going to say the same thing as Adam – SMART is nearly worthless. That standard was created without first doing proper feature discovery to figure out what kind of data would actually be useful in predicting mechanical drive failures.