I manually locked the FSS by disabling the autolocker, moving the temperature to where it was when the FSS was previously locked, then engaging the loop.
Looking at the FSS time series, it seems as though the temperature never actually reaches the top of it's range.  When the FSS sees flashes, it triggers some FSS_OSCILLATION state in the PSL_FSS guardian, which ramps up and down the common gain in an attempt to unrail the PZT.  It's possible that these common gain ramps were too small, happen too quickly or jerk the temperature around too quickly.
I adjusted the FSS_OSCILLATION state so the gain goes down to 0 dB from 28 dB, then ramps back up.  Unclear if this will actually help, probably would be smarter to tune the autolocker settings to slow things down around the flashes.