This is only as far as I can remember as it has been a while.
If you had transverters etc plugged into the back of the 901/902 and the acc socket, then generally for the transverters the accessory plug removes power to the heaters on the PA valves so they are effectively isolated, RF power for the transverters came out of the driver valve RCA socket. There is a diode detector on the output stages providing a negative voltage that is fed back into the accessory ALC connection.
If you are driving the FL2100Z then the accessory plug has the heater link for the 901/902 PA valves and the RF comes out of them into the FL2100Z. If you look at the output side of the FL2100Z circuit there is a diode detecting the level of output power and feeding this back to the ALC circuit which then can reduce the previous RF stages in the 901/902 to maintain a reasonable power level.
So going off these two scenario's, I basically used a low level diode detector from the driver to feed back to the accessory ALC pin.
This seemed to work well and I believe would not cause the concern that G6Tanuki mentions above as any residual carrier would also be attenuated by the ALC circuit in the 901/902. Doing reduction by reducing the level of audio it would also be possible to cough/sneeze in the microphone and get out a lot more power.
As to power levels I know on 14MHz I can get 100 Watts and I think up at 10m I am getting around 90 Watts, but I have only got a DAIWA CN-101 power/SWR meter so how accurate that is I would not know.
I joined one of the old 901/902 groups again last night and was able to find the very simple circuit I used.
The diode detector was built into an RCA plug the 500 Ohm pot was at the side of the 902 and then the cable fed back to the acc socket.
The CR time constant may want looking at but it is simple to try.
Adrian
p.s. My 902 basically sits here like a large paper weight due to restricted antennas and prying neighbours.