Why do you think he is oversampling Republicans? He says on his website that he is weighting by party based on a six point plus spread for Democrats (which is too high).
I just crunched some numbers and am beginning to suspect that last night had to be an outlier (which, statistically, happens about 5% of the time). Using Microsoft Excel, fancy math, and some guesswork, I determined that Obama never polled higher than 50 and lower than 47 on any night from 9/13 to 9/22. But to get the result he got today, he had to have pulled 53 last night. This was after he pulled 47 the night before. No way he jumped six points in one night. [Note: I don't actually know what the nightly snapshot numbers are -- this is educated guesswork using Excel. I do know that these nightly numbers on my Excel chart are consistend with Scott's reported results for this period.]
Also, it isn't that McCain had a bad night yesterday if you look just at his numbers. Again, using educated guesswork and Microsoft Excel, I figured that from 9/13 to 9/22 McCain's nightly numbers ranged from 45 to 49. Last night he polled 46, which is pretty much in the fairway of what he has been polling.
Using common sense, nothing happened yesterday that would cause a bunch of undecided voters to switch to Obama, especially when McCain apparently didn't lose any support at the same time. So, I'm guessing that Scott just had a bad sample last night.
Gallup has D +9, ABC D +10 party gap in their polls. This was consistent with Rasmussen’s numbers throughout 2008.
Currently Ras had D +5.5 as his baseline, but this involves the RNC bump, and may be too optimistic. Definitely, nobody (well, may be Battleground) has it that close.