IIRC it wasn't truly stable, just stable enough to sometimes finish a Geekbench run before crashing.
It's slightly arbitrary, though; the real limit right now is the voltage limit set in firmware of 1.0000V. I think we could get a stable overclock to at least 3.2 or 3.3 GHz with more voltage.
Right, it does come full circle by half.
it can be done higher if voltage is added further. Maybe with certain temp control can achieve 4GHz or more