There's one pretty big flaw with this approach... For certain searches that do not have "millions of results", you get completely unrelated results.

If I search my name then the results are for names similar to mine, but not actually my name. This makes it completely useless for searching my name. I would think that there are many searches with this problem.

I think there needs to be some kind of weighting system used that dynamically decides the cutoff point. One million is a huge over-generalization for all search terms.