John Shaw had an interesting post on his blog a few weeks ago about ETTR. In fact he called it exposing WELL to the right.
He looked at exposures that appeared to be overexposed on the original histogram. When those raw files were loaded into his imaging processing software they were not overexposed at all. He did a test of his camera and determined that he could go 1, 2 or 3 stops more than the histogram indicated for overexposure. Seemed to indicate that the camera histogram may not be as accurate as we would like.
His point was that by testing his camera and determining that he could go past the point of histogram indicated overexposure, he still had an image whose highlights were not clipped and he could get an even better image (less noise), Seems to be the same thing several replies here are indicating.