Here is some investigation on photodiodes saturation for an unlock of Friday provided by Julia and Diego.

The first plot is showing various signals before the unlock. There is a low frequency oscillation (0.6 Hz) that does not help. But if you zoom a bit near the unlock (second figure), and check the top left plot, you see saturation on the B2_8MHz signal (see the green arrow).  Of course, you may think that the fast ADC is not saturating (counts around 5000, far from the 8192 ADC limit), but here the VGA gain is low and this is the preamplifier that is saturating.

Checking a bit further the sample data (blue traces of the third figure), we can see nice saturation at 1 mW, that is a large RF signal compared to the 3 to 4 mW of the DC signal. The last figure is showing the sample data for a good lock of a week ago, were the RF signal was not saturating, although it was close to the 1 mW (the DC power was 2.9 mW).

This suggests that it might be good to reduce the (RF) power on the B2 photodiodes, may be by a factor 2.