Reading this comment, I thought that the 2,000 to 5,000 year claim was complete rubbish. Turns out that it's not, which is a bit surprising - it's supported solely by computer models at this point, not genetic evidence, and is probably contradicted by groups that were not interbreeding at the right times, but still.

http://tedlab.mit.edu/~dr/Papers/Rohde-MRCA-two.pdf