The number for 18-34 y/o voters looks about right, at least for the LV model (RV looks a little bit oversampled). I'm wondering where they got the absurdly low 35-44 y/o number though. It's low by at least 4%, I'd say. The big trick is with the D/R breakdown. 9% D advantage, which mimics the Operation Chaos/ACORN tomfoolery, but is at least a 6% departure from historical breakdowns from within the last two years, and is even further afield from the post-voting partisan breakdown in the 2004 election.
One thing that bothers me is that I've been hearing of a number of people, such as our own FReeper Italianquaker, who report that they were called by a polling outfit, but were hung up on after they identified as Republican, conservative, McCain supporter, etc. Hung up on! I can see maybe weighing certain respondants out from your data set to fit your particular blend of special sauce, but hanging up on people? Refusing to even gather someone's data after they identify a certain way? IQ isn't the only FReeper who's reported this, and I've heard anecdotal evidences elsewhere to the same effect. Hanging up on someone strikes me as a singularly incompetent way to conduct a poll, to be sure, and indicates to me that there is more than just some past-the-expiration-date special sauce going into some of these polls.
How about some behind-the-scenes threats? Pressure from the Obama campaign? The Thugocracy may have started a year ago.