Looking at the data as collected, a quick mapping of it shows an interesting linear relationship with the three loads you used and the outputs.
If you get the first three rpm (73 -162) and the 35R load, the next three rpm (245 -321) and 75r load and next three rpm (426-1065) and plot the points, and almost linear relationship seems to exist.(kicks up ever so slightly after 700rpm) This is not what I would have expected. Power should have gone up much faster. I assume your high reactance with the iron core is to blame here for it to not go up as a square of the rpms.
If you were to use a pwm device, it may be possible to use a simple linear relationship from your tacho to pulse width, and achieve a close approximation to the best you are going to get anyway. The delta switchup could be left until 800rpm or not at all, if you continue to relieve the load on the star as rpm increases beyond 700 rpm. This could simplify the switching system (none at all) and not sacrifice much in performance (on these figures anyway).
Look forward to your endeavours in this.
Just a quick note, If you use power factor correction for the dc input to the PWM, then at the 0-200 rpm range at least, even a good pwm (80khz,500w etc) would range from 70%-87%.as the voltages here will be pushed up to your 350 or so volts. As the voltage generated gets up in the 200v-300v range, then perhaps we are looking at 95%-96% If we then place the inefficiencies of the PWM on top of those efficiencies, we are starting to get a fair bit of inefficiencies built up by the very devices we are using to drop the inefficiencies.
Worse still the PFC will be going pearshaped just when you really want the most help........... A PFC/PWM combo will probably come in at around 87%, by my guestimate.....but at less than 200rpm, then maybe 75-80%......life a bitch sometimes just when you want it to be 100%..
Well that matches it to the drill press, now for the blades to complicate things.....
Keep it up Commanda..........oztules