I haven’t been able to stop thinking about this. pretty excited to understand it better.
I did another version of the power curve comparator on desmos. parameterized it in terms of NsMp cell counts. and added a “gearing” component.
In the default comparison, 10s6p (blue) and 12s5p (red) the power curves are the same. Red has more top speed but the torque/amps continues to drop as the rpm/volts (dutycycle * volts) goes down. keeping max power the same. (!!!)
bingo light going offf for me. 
That power curve staying the same is why:
Now,
if I were to gear it down to the same top speed. I do gain torque. but only where motor amps previously limited blue. and only where it stays under the max power curve. ( red is now virtual amps/torque relative to blue. but really it’s gearing )
so higher S count’s extra speed can be traded for more torque but only where previously limited by motor amp max. it can’t exceed max power. This is the tradeoff I was thinking of. but now I can also see the max power curve and how it limits the effect of that trade off.