Actually nothing all that odd happened. J. J. is as usual cherry-picking results and doing it badly. In New Mexico (which is always notoriously difficult to poll), the polls had an average of him with 55%, and he got a little under 57%. In Arizona Obama got a little under 45%, and the last polling average had him at 46% (and of course why undecideds might break for McCain in Arizona is a little obvious). Nevada's the only one where the polls were off by a significant margin, where Obama averaged 50% and he got over 55%. However the last poll average also showed 6% undecided, and with the swing Nevada took and the economic conditions of the state it's not too hard to simply seeing the undecideds breaking heavily, and with the dynamics of the state using a turnout model based off 2004 in 2008 would be somewhat inaccurate. I'm oversimplifying quite a bit obviously, but these are far more logical explanations than "The Bradley Effect causes black candidates to underpoll in states with lots of Hispanics."
Also look at Texas where Obama polled at 41% with 5% undecided and got 43.6% (aka the undecideds broke almost 50/50) and California where he polled at 59% and got a little under 61% (with 3% undecided.) So this "inverse Bradley Effect" thing with Hispanics basically requires the same total random occurrence to be believed in.
Do you attribute Reid and Bennet overperforming their polls to a similar sweeping of the undecided vote? In any case, Reid and Bennet aren't black.