Some commercial eBooks seem to have OCR mistakes, but this is rare with new titles. The most common mistake I see is a word broken where a hyphen should be, or a hyphen inserted where one shouldn't be. I see this even in new books. I think this happens when a human goes back and manually corrects the H&J on the layout for the physical book, and the script they use to convert to OEBPS doesn't handle it correctly.
I imagine there's at least a couple of employees at every big publishing house who go back through and touch up a book "incorrectly," thinking they are making the final product have better quality.
I'm also betting that very little proof-reading is done on most eBooks, with the thinking that the proof reading for the physical books was enough.