Oops, in post #5, I messed up the operating description. The screen feedbacks will cause the driver plates to adjust -voltage- to get mu times the input delta V (at the grid1s) to occur on the screens (from the output tube plates). Just like a triode, delta V screen over delta Vg1 stays a constant, mu. The grid 1 inputs do set a driver current partition in the differential stage, for 1st order output drive, but the screen grid feedbacks will fine tune that ratio to get the V ratio from input signal to final output V signal in the ratio of mu. When mu is satisfied, the plate current then stays constant, otherwise it changes the current ratio until mu is satisfied. Using gyrator loads for the driver plates makes for very high gain in the control loop. Driver Ri constrains the loop gain possible.
With the gyrator driver loads, the current in the two driver tubes will stay near equally partitioned, as the grid1 input signals and the screen feedback signals swing in opposite directions by mu ratio to keep the driver tube currents near steady. The driver plate voltages will however be swinging to whatever voltages are required to get the output tube to behave properly for mu enforcement. Made possible by the high Z gyrator loads. Essentially a tube Op. Amp controller. Adjusting the final output until the feedback nulls the input (with a mu factor N Fdbk attenuator to get mu gain plus any resistive attenuation factor). The driver mu does change slightly with input voltage here, so you get a triode gain result.
Lowering the gyrator Z (to some resistor instead) will give you the same result as a loaded triode, with increased 2nd harmonic.