EDIT: I should have mentioned that rotating the polarization before the EOM in order to minimize the error signal offset did not have a noticeable effect on the bulk of the low frequency noise. It did, however, make the coherent-looking peaks from 100-600 mHz go away, though I have no materials to support this.
Also, by placing a PDA255 after the initial beam splitter and demodulating it at the sideband frequency, I observed that the RFAM noise is flat well above the frequencies where we see the excess noise. This isn't purest way to rule it out, but I think it is convincing: if the noise we see at low frequencies is from RFAM, we would see it extending to higher frequencies.
Below are the (delayed) results of Friday's oscillator noise measurements, calibrated to gyro units and plotted alongside the gyro noise. The traces are: