Last night I took a closer look at the LSC analog signals to find which components are making the glitches.
I monitored the RFPD output signals and the demodulated signals at the same time with an oscilloscope when the PRMI was kept locked.
Indeed the RFPD outputs have some corresponding fast signals although I only looked at the RELL11 I and Q signals.
(REFL33 didn't have sufficiently a high SNR to see the glitches with the oscilloscope.)
I will check the rest of channels.