For 4 CPU systems on my Linux machines with a BOINC setting of 75% CPU at 100% time, if you look at a CPU usage monitor, all 4 CPUs appear to be running at 75% of the time as the linux scheduler rotates the workload to keep all cores roughly equally busy and at the same temperature. Can only speculate that the linux schedular is moving work around much faster and more efficiently than the BOINC algorithm that keeps a CPU busy for 75% of the time with all CPUs busy. I would expect Windows to work similarly.
With the linux systems here a different approach has been taken, that of nominating a maximum CPU temperature and decreasing the CPU frequency, if necessary, to stay within that limit. A user program was written to automaticaly accomplish this. As expected, some of the faster systems here with a greater number of CPUs cannot be run with all CPUs at maximum turbo frequncy during the very hot Australian summer, as the office in which they run is not air-conditioned. As a result they produce more results in winter than summer for a given time as the governed frequency is higher in winter. BOINC is set to run all CPUs at 100% of the time.
CPU power efficiency decreases with increasing CPU temperature. On the Intel i7 3770 here , for example, the increase in frequncy from 3.7 to 3.9 GHz (max) produces a massive amount of heat and temperature increase compared to the same increment in frequency at lower frequncy settings. The hotter the CPU, the greater the junction leakage.

Yes, absolutely. Go with % of CPUs in a multiple that will allow you to drop cores/threads to keep the temps in whatever range you need. Cycling CPUs between 100% load and 0% every second will lead to thermal stress and premature failure.

I'm using 50% CPU + 60% CPU time. On my Lenovo i3-8145U CPU - 8GB Ram - Win 10 1909 PC, it gives me good task throughput and no lag whilst using other apps concurrently like Edge, Mail, etc.