Today we vented the chamber and obstructed the cavity internally to measure the open-loop noise on the PDs themselves. To account for noise in both PDs while still rejecting anything that was common-mode---things like laser noise that are suppressed by the primary loop and absent from the gyro signal---I took the demodulated outputs of the PDs (i.e. the open-loop error signals) and subtracted them with an SR560. I then relocked the cavity, measured the current gyro noise spectrum, and took OLTFs to obtain optical responses with which to calibrate the error signals into gyro noise. The noise is far too low to explain the excess LF gyro noise:
I also measured the error signal spectra individually. Both of them are at about the same level as the differential noise, so there isn't a huge common-mode component.
As far as I can tell, we still haven't found the source of the blasted LF junk.