[morzh]

Dale,

Honestlt, if I use a compound (rarely), is try to design thermal path such that it will not fail should the compound be measured a trifle incorrectly. That is, I try not to design close enough to critical point.
With or without compound, I try to ensure that, while within the specs of the ambient, the device never gets too close to it, even if it is still within the safety zone.

For instance, I use Vicor power supplies (yes, I understand we are talking of a bit different power levels) that are 400W bricks, that specify 80-85% efficiency. Which is, to think of it, dismal. It means that, should i put out 400W, I will be dissipating up to 60W at the thermal plate (this is where the power transistors are mounted). Given that our device is only specified to 250W power, this would be 38W, but even this is a lot. And it is (not through my fault, the design is legacy and the rest of the team is not willing to change the mechanical design) mounted to a 3mm or so ANODIZED aluminum plate, which purpose, BTW, is not being a heatsink. It carries other stuff.
There is a fan in the device, but it is fairly weak as we do not want lots of airflow, as it is a fiber splicer and works with (can work with) fairly fine fiber which will oscillate because of that flow, and you will not be able to splice.
And the response of the mechanical engineer was "I do not want to change anything" even all that I was asking for was, let's mask the plate under Vicor during anodizing and iridite it instead. Let alone the fact that the Vicor's plate has to be electrically connected to its metal mounting plane to use it for EMI puwposes, if it is Grounded (and I also found out it was not). Iriditing would ensure botyh thermal and electrical connection.
So, eventually I was able to make them to mask that part of the plate.
But the plate, the way it is, is not a good heatsink.

So I went and measured the temperatures right at the Vicor plate, where it meets the mounting plate.
What I found out that, somehow, without reflecting it in datasheets, Vicor MAXI series is quite a bit more efficient than what they claim. Obviously, there has been some improvement over all these years (the thing is getting close to 30 years, and is very unique amongst the Telecom Bricks by it's Trim range, from 10% to 110% of the output voltage, and in reality this is even better).
So, now I know we have good margin. But I would never approve the newer designs (not that they would listen; after all the devices are precision mechanics and my role is supportive, controlling it all, the secret sauce is in the mechanics and the process) explicitly, had I not known that I had had a very good temperature margin. That said, while some compound is good (Vicor does not exactly has the smoothest thermal plate surface; BTW if it does, it is the tell-tell of counterfeits), its less than correct application will not put the brick in danger zone, even in still air.

I do know about price/reliability design compromises, but, even within the budget, I do not go by "OKI, it is STILL not in the danger zone, fine". I go by "OK, it is 50 degrees away from the danger zone even if all goes wrong."