To fix this, I am going to put a 6.8uF cap in series with R30 in the MC board, which is part of the crossbar switch where the IN1 and IN2 get summed. This should AC-couple the output of the IN2 slider before the summing node.
MC board is out, so don't be surprised that the MC isn't locking.
MC board is back in place, MC is locked.
If I disable all of the AO path bits of the CM servo (disable switch, and also gain slider to -32dB), and then move the MC IN2 slider around, the MC does not get an offset anymore (as seen by reduced transmission and increased reflected power), so I think the DC coupling is working. I do lose lock of the MC if the slider goes above ~22 dB in this situation, but I don't see any effect before then, whereas we were able to see a steady increase in the reflected power as we moved the slider around last night. So, it seems like things are good with the DC coupling of the IN2 slider.
Here are some photos from before I modified the board (front, back, and zoom of the area I was working in):
And here is my modification, putting the 6.8uF cap in series with (a new) 2k thin film resistor, in the spot for R30:
The board is https://dcc.ligo.org/DocDB/0004/D040180/001/D040180-C.pdf
[Edit, 20140721: It looks like this is actually D040180 rev B, not rev C. —Evan]