Making judgement on thermal output based on temps? Come on.
Despite your tone, you raise an interesting question. Empirical data rules the day.
I hooked up my Kill-a-Watt, and based on the current power draw readings, I estimate that my EVGA P2 750W is sitting in one of the worst parts of its efficiency curve, giving it about an 88% efficiency. I tested two settings, comparing the package power readings of HWiNFO64 to the readings on the Kill-a-Watt adjusted for PSU efficiency. Bear in mind that board + VRMs will account for some discrepancy. I ran y-cruncher since it pushes hardware pretty damn hard.
3.6 GHz, max fans, no power saving, 1.35v vCore/DDR4-2133, 1.2v vDIMM:
Package Power @ load: ~95W
Package Power @ idle: ~17W
Kill-a-Watt @ load: ~229W
Kill-a-Watt @ idle: ~114W
Same as above but 4 GHz/DDR4-3200, 1.4v vDIMM
Package Power @ load: ~125W
Package Power @ idle: ~24W
Kill-a-Watt @ load: ~242W
Kill-a-Watt @ idle: ~110W (?!??!)
The suggested jump in package power @ load from 3.6 GHz to 4 GHz was ~30W, but the Kill-a-Watt showed a difference of only 18W between the two settings. The Tctl delta was ~20C (max Tctl @ 4 GHz was 86C, max @ 3.6 GHz was around 65C).
So I don't know what to make of that. Something is fishy here.
Also, HWiNFO64 records a CPU temp from the motherboard. According to that sensor, the chip never went past 42.5C @ 4 GHz. Even given the 20C offset for Tctl, that's a sizeable difference in readings between Tctl and the board's CPU temp sensor.
edit: Also, my Kill-a-Watt is reading insanely low power usage for the clockspeed and workload involved. y-cruncher is (for Ryzen) an AVX workload that pegs all cores, hard. It has produced the highest Tctl values of any program I've run to date on this admittedly-young system. Can it really only be using a total system power of 242W? The package power readings seem to be too high for the overclocked system . . .