Anyway, back on topic:
I think the M295X is going to throttle no matter how much thermal paste you put on there, but I'd LOVE to be proven wrong, of course. Until then, I have no intention of ripping apart my iMac. I have numerous adhesive-strip kits to put back the 2012/2013/2014 screens on, so that's not a concern.
I totally forgot to reply to this part the first time around, so here is try 2.
You are right about the M295X throttling no matter what. It has PowerTune and Boost enabled, so the clocks are going to dangle below the maximum advertised speeds depending on the type of workload (the more the workload is tuned for the chip, the more power the chip will sap, and the more often it will drop in clocks, for example). Nvidia advertises a minimum clock rate, so you'll see the clocks "boosting". The same thing, just two sides of the coin. A power budget of 100-150W is trifle for this chip, so PowerTune steps in to control the power used by the chip by changing clocks and so on.
Now that I see at least some people are reporting >100 Degrees Celsius on that chip, I think perhaps there is a bit of thermal limitations in some cases. It would be interesting to do a poll to see how hot people's chips are running when subjected to a fixed workload (say, OCCT or Furmark) on this forum.