Summary: they averaged the last round of polls for a bunch of swing states, scaled it to adjust for the undecideds, and compared it with the way the vote actually went. There's nary a hint of the Bradley Effect -- in no case did Obama do significantly worse than the polls said, and he actually did considerably *better* than the polls indicated in a couple of those states.
So the general verdict is that you can put a fork in this long-hypothesized political force: in its biggest test, there's no evidence that it had any noticeable impact.