Yes; this is one of my ideas for things to try...
Make the noise budget of the circuit. That's the only way to reduce the noise rationally.
Attached is a current gyro noise spectrum, taken by measuring the noise in the actuation signal to the AOM VCO. By comparison with the previous spectrum, it is clear that the noise has gone down by a factor of ~2 after the modifications we made to the CCW PDH box.
The proposed mechanism for this is that when the CCW loop acts un-ideally, there is some residual common-mode noise that is corrected for by the CW loop alone, which results in an apparent gyro signal. To test this, I reduced the gain setting of the CCW box from 3.5 to 1.5 and re-measured the gyro noise. The result above ~200 Hz makes sense, as the noise goes up in this region by a noticeable amount. Below 200 Hz, however, the noise seems to have gone down. I'm not sure what to make of this at the moment, but I have some ideas for things to try.
For those wondering, we are looking at the gyro noise in this pathway because it is more practical to do it this way right now. The linked entry above shows that the noise in the AOM feedback signal is actually lower than that in the transmission PLL readout, and we have to get some factor of 100 lower until we expect to see benefits in looking at the PLL noise. Also, this rules out calibration errors and the like from things shifting around in the PLL between measurements.