Here is a summary of what we tried just now. To cut to the end of the story, we finally found that the 95MHz signal has disappeared from our PD. We don't know why because both beams are still on the photodiode.
Okay, back to the begining. We started by looking at the 95MHz signal on a spectrum analyzer and trying to maximize it by moving the beams on the PD. We weren't able to get the signal above ~ -68dBV. We increased the laser power on the PD to around 50uW, and we had a 95MHz signal of around -62dBV.
At this point we connected the signal to a mixer, and used a Marconi as the LO for the mixer. We set the carrier freq to match the AOM freq, and then set the Marconi to Ext FM, with deviation of 100kHz. We connected the output of the mixer to an SR560 which we connected back to the FM input of the Marconi. We were looking at the output of the mixer and the output of the SR560 on a scope. After quite some time we hadn't seen anything that made a lot of sense. We re-measured the 95MHz signal on the spectrum analyzer, and it had gone. We checked that both beams were on the PD, and that blocking either caused the voltage out to drop.