I work for Intel but I don't have any inside knowledge of this feature.
BUT, I have done some empirical tests. First, I uplugged the fan while running a benchmark (Prime95) and noted the benchmark result while monitoring the CPU temperature. I posted the results of that experiment here about a year ago. Unfortunately, the results of this test left too big an error on what temperature correlated with which benchmark result because the temperature rose faster than the benchmark could complete. So then I put a rheostat on the CPU fan and dialed the fan speed, let the CPU default to a fixed temp and then benchmarked it. What I found was that there was a fairly steep onset to throttling. Around 82C (number from memory, I did this about a year ago and I can't seem to find my notes), the CPU performance pretty much dropped off completely. Up until that point performance was fairly constant (within a percent or two).