The IMC remained locked over the whole weekend.
It seems that, somehow, increasing the current of the ML pumping diodes, reduced the amplitude of the glitches on EOM_CORR (see plot 1 and 2). 
During the weekend the correction did not reach the +- 1V that makes the IMC unlock.

The only unlock was on the 25/01 at 06:12 UTC. It does not have the shape of the usual fast unlock. The correction of the ML_PZT starts to drift about 1 s before the unlock (see plot 3). This drift is not explain yet.