Beyond that, however, the poster needs to learn about ROC curves (http://en.wikipedia.org/wiki/Receiver_operating_characterist...) – for just a moment, let's assume we're genuinely only hiring "A" players – then the challenge is a signal detection problem, pure and simple – sorting the signal from the noise and avoiding both false positives and false negatives.
While I have huge issues with the post's method, it's clearly only focused on avoiding false-positives – avoiding hiring someone who's bad but is sold as good/great. Unfortunately, there are real issues with false negatives as well – you may miss out on a fantastic candidate, first and foremost...
But, of course, decisions aren't that simple. Anyone who understands decision making knows that decisions depend highly on context – including the context provided by the other options... and if you systematically fail to account for eliminating some number of options, you may well be (foolishly) limiting the range of options and the context of your decision. In short, "A players" are relative...
And that idea goes even further. Many stellar employees, innovators, coders, etc. don't start out that way. Sure, some do, but those aren't the people applying and giving you references.
Your job, when hiring from a resume pool, is precisely to find those diamonds in the rough – those false negatives.