I'm pretty sure most polling adjusts their pool to fit their model of likely voters in spite of whomever they interview.
Here's an article from Pew on their methodology for party affiliation (from July): http://www.people-press.org/2012/08/...lection-polls/
I guess they don't take in account party affiliation when extrapolating, but they do note that about 8% more of the national population identifies as Democrats. And that might be higher in Penn. So a D +10 isn't all that dumb.
I also imagine that as elections become eminent independents identify with the party with whom their presidential choice resides.
All this said, if you're trying to dig through the numbers and find a silver lining for Romney's chances, I think that's pretty fruitless and wishful.
Nate Silver's 538 methodology can be found here: http://fivethirtyeight.blogs.nytimes.com/methodology/
It's interesting stuff; if you're curious.