I finally realized where the issue was.....
There was a spring as part of the tension mechanism, put in because the pulley sets don't exactly take the same length belt on every step. It took up the difference.
It is a very short, very stiff coil spring, but it was effectively limiting the belt tension, and hence the max power. More pull would deflect it and tend to hold tension flat. The actual max tension was less than it should be for some mechanical reason. The spring itself didn't limit the max, but limited the maximum on certain pulley steps if it was set correctly for other steps.
The compensating spring was not at the same deflection for all steps, and hence was not giving the same tension.
the alternative is to adjust for each step, which is a big hassle. However that is what I have to do now. I need to count turns of my freshly made adjusting nut with thumbwheel, to get the tension up to the approximately 4% stretch which does the job. The result is that I can get a reasonable tension which takes full motor HP for the higher belt speeds.
I clearly would NOT have been able to do that without polishing the pulleys.....tension made far less difference before polishing.
I am trying to come up with a provable spring method..... possibly belleville springs would do it, as they don't change force over their entire deflection range. All compensations within their range would have equal tension, unlike the prior situation.
Since the tight side tension goes up, but the slack side tension reduces, it may be possible to get a tension setting which will be constant for all settings, and also be high enough to allow full HP where needed without slippage.