I hope this solves your problem. The reason the LM393 will not work with high input common mode voltage is that is has a PNP differential input stage, and each input also has a PNP emitter follower. This means that the input threshold voltage (from your voltage divider) has to be low enough to allow two series PN junctions to turn on, in addition to allowing some headroom to allow at least a couple of tenths of a volt for the tail current source of the differential pair. This adds up to, as I said, around 1.5v. BTW, this spec goes to (Vcc-2v) at 0 deg C.
I also wonder about the reliability of the printer that you mentioned. PN junctions have a negative temperature coefficient (~ -2mv/deg C), so as long as the comparator stays warm, the VBE drops will be relative low, and less headroom is required.
I suspect that you could buy some cold spray and hit the comparator with it and it would fail reliably. :roll: I can't explain why it worked when you took it out of the freezer.
BTW, another potential fix is to power the LM393 from a higher supply voltage - but you probably don't have one.