I checked the spectrum of ALS_DARM error signals with (red: spectrum, blue: RMS) and without (pink: spectrum, cyan: RMS) locking ALS_DARM loop (fig1).
DARM error signals becomes large when locking ALS_DARM around 30 Hz by a factor of 3 and the bumps around 30 Hz are dominant source of the RMS above 10 Hz.
Since DARM loop UGF is designed at 20 Hz, 30 Hz bumps probably come from the gain peaking due to poor phase margin at UGF.

Figure 2 shows an example of the modification of OLTF.
Left graphs show the gain and phase of OLTF in current design (UGF is 19.6 Hz, phase margin is 32.2).
Right graphs show the gain and phase of OLTF added lead filter (z:15, p:30, g:0.752).
Phase margin around UGF becomes 50 Hz if we added lead filter.

So, redesigning of ALS DARM loop might reduce the RMS of feedback signals and help the stable lock.