I'll piggy-back on the above statement, and add a bit more.
For a moment, imagine this as an efficiency issue. You put in 100 watts, 60 is converted into heat, 40 is converted into potential differences, and you complete a task in 10 seconds. The same processor may be able to take 50 watts in, convert it into 28 watts of heat, and complete the task in 19 seconds. It doesn't sound like a lot, but you have half the energy input, less heat, and less than half the time to complete the operation. Likewise, you could pump 200 watts in, get 130 watts of heat, and take 6 seconds to complete the operation.
All of this is to say that energy input, computational output, and thermal conversion are not uniformly linear in a processor. Simply waiting until the thermal limit is reached, and allowing the processor to throttle, isn't likely to be the most efficient use of energy...ever.
This is why you see people undervolting chips to get them to run cooler with minimal performance losses. It's also why I look at the 7000 series from AMD and really ask whether using thermal throttle limits to get performance is a long term viable solution. I'm sitting on a 3930k which used to hit about 3.8 GHz, under water cooling. A decade on and it's down to about 3.4 under the same cooler. That's not a huge drop...but I also never ran the thing stupid overclocked.
My current driver 3700x is slightly undervolted, and the frequency is slightly upped. It isn't golden chip levels, but I also want to run the thing for another decade. I'm...reticent to say that the 7000 series will soldier on that long without some falloff...that may make my 3930k tame.
