The question is whether there is a headroom in overclocking. Jensen claims that there is.
There is always at least some headroom.
Even if NVIDIA's GPU Boost was perfect at extracting all practical, stable, performance from a given power limit and temperature, power limits can go up (allowing more voltage along with it) and temps can go down.
I just hope the miner was sincere about the card not running high temperatures, as mentioned I did try it out and it works very nicely, fans spin up without noise and literally, there was not a single speck of dust on it. He also had a 100% upvoted rating on the used market.
Most recent cards have very conservative limits in what you can do to them without hardware modding. There is no reason to think a Pascal part used for mining would be in any worse shape, except perhaps when it comes to fans, than one used for much of anything else. Peak practical mining performance, with no regard for efficiency, will mean the card was run with 100% fan speed 24/7, and was lightly overvolted, but saw essentially no thermal cycling and fairly cool temps. Peak mining efficiency means the card was very cool, probably undervolted significantly, and also had minimal thermal cycling.
Thermal cycling and the resulting mechanical stresses on solder joints are what's going to kill the majority of GPUs. A card used for gaming a few hours a day is likely to see many cycles with a ~40C temp delta and may see a daily peak delta of 60C or more. A card that is mining either sits at a fixed temperature target (to ensure predictable hashrates) or varies slightly with ambient temperature. I have cards that were mined on for twenty or thirty thousand hours which saw fewer lifetime thermal cycles than I can get in a weekend of playing Elite: Dangerous.