I found 2012: As expected, it was D+6: 38/32/29 -- almost exactly the same as 2008.
And in 2014, it was R+1: 35/36/28 -- in exit polls conducted for House races.
So, the question remains: given the dynamics of voter enthusiasm, is oversampling that exceeds the difference observed in Obama's first election the correct value to use for this election?
I found this resource for Presidental election:
http://ropercenter.cornell.edu/polls/us-elections/presidential-elections/2012-presidential-election/
On the right side, you can choose Presidential elections back to 1976. Click on the link labelled {year} Group Voting, and then scroll to the bottom. You'll find a category for Party Identification, and the first column of numbers is the total for that party.
An interesting piece of data: in 1984 it was D+3: 38/35/26, but Reagan nearly swept the Electoral College. It was also D+3 in 2000, when it came down to one state (Florida) in the Electoral College.
There were more democrats than republicans voting in 2012 - that is why Obama won, Sherlock.
In fact, the percent of republican representatives in the House exceeded their vote by a fairly wide margin, which is why the democrats complained about republican gerrymandering, again.
There is more straight ticket voting now than there has been in two generations. That is why the republicans were panicking about losing both the house and senate just a few weeks ago since so many voters are straight ticket voters now. Since the polls are much better now, republicans should still keep the house, but keeping the senate is going to be a bit iffy.