Since 42866 we know the ML to Slave frequency loop has an effect on the IMC lock performances.
An explanation could be that the injection band has reduced from few MHz to few 100 kHz and makes instable the slave injection overall loop.
the process of optical injection insures the slave laser gets the frequency of the ML up to the injection band (~ few MHz). The analog loop provided by the "slave" rampauto insures that both the ML and slave cavities are frequency closed to each other and this guarantees a low phase noise of the slave laser beam wrt the ML beam.
From the loop POV it means there is the rampauto loop (ugf=100 kHz) in serie with the optical injection (ugf ~ MHz)

A decrease of the injection band to 300 kHz (for ex) could make instable the overall injection loop.
The injection band goes like sqrt(P_SlaveLaser/P_MasterLaser). A reduction from MHz to 300 kHz means the ML_power is reduced from 1W to 100 mW.

on 6/04/17 we have measured the injection band to be 2.35 MHz (37175) see plot1, error signal central part width = 2.35 MHz) by frequency scanning the Slave wrt th ML
on 25/09/18 we made the same meas. with a somewhat different shape for the error signal (see plot 2, The modulation for the LB is 14 MHz). The error signal central part is higher that the LB bands parts (this is typical for cold cavities but should be the opposite with the injection lock)
The error signal width is 116 kHz instead (would mean the injection band has reduced??) and with the feature of cold cavity.
 

It maybe worth to scan the slave laser with slower speed and check again.

The measure of the TF from the EOM up to the slave error signal could be usefull too to detect issues with the EOM and/or with the optical injection