I'd love to find some rigorous way to measure it, but (as of six months or so ago) when I sat with two phones, held down a button, and waited for a tone to come out the other end, the delay was observably longer with a GV-to-GV call than with a GV-to-cell call, which took longer than cell-to-cell.

I find myself saying "no, go ahead" or talking over the other person far more often with GV than not. It's certainly usable, but I avoid GV if the call quality really matters.