I'm wondering if the increase of the phase noise at high frequency when the cavity is locked is due to the fact that, when the cavity is locked, the frequency changes of the master laser are very large ~ MHz.

Possible tests to check this hypothesis are to damp the mirrors (sending the PZT correction signal to the mirrors, upon filtering) or to excite the mirror oscillations, to artificially increase the laser frequency changes. 

A related question: when we compute the residual RMS phase noise between the main laser and the auxiliary laser we integrate down to 100 Hz. Maybe the 1 Hz region is dominant with respect to the high frequency region, and thus we should solve in any case this problem of  the main laser frequency changes in the Hz region.