Nulling the slow actuation offset fixed the issue. Now PMC is back to normal.
The reflected beam on the CCD was quite symmetric (it looked very TEM00 mode !) for some reasons, I somehow suspected the mode matching to PMC.
One possibility I thought of was the laser temperature because it could change the laser spatial mode.
So I looked at the slow actuation offset on the FSS screen and found it was at -4.0 which sounds somewhat big.
Then I zeroed the offset by the slider and relocked PMC.
Then the spatial pattern of the reflected beam became usual (i.e. junk light looking) and the transmitted light wet up to 0.83 which is normal.
The PMC is losing power.