Thanks to commenter virginiacentrist for directing me to this useful Kos diarist post. I don't I buy his/her bottom-line prediction, but the point about regional vote shares is right on target. From the diarist's summary of the polls, it looks like the necessary (but not sufficient) condition for Obama to do well tonight is for Philadelphia and Southeast Pennsylvania's share to get into the mid-40s and for the Pittsburgh/Southwest Pennsylvania share to stay under 30 percent. If we get below the mid-40s for the Philly region and above 30 percent for the Pittsburgh region, Hillary's probably going to win by double-digits.
As before, I say necessary-but-not-sufficient because Obama would still need to win by a healthy margin in the Philly area (probably double digits) and not lose by too big a margin in the Pittsburgh area (under 25) to have a good night even if the reginal vote shares go his way.
Update: Mark Blumenthal has no firm conclusoins but some good closing thoughts about the behavior of undecideds and whether there might be a late break toward Hillary.