Are the decoupling caps on +12V right near U1 & U2?
Can you see anything unusual on U2A-1 with a scope when a triangle wave input signal is applied to the input?
Since U2A isn't switching correctly, I'm thinking there's some spurious coupling to its '-' input. A quick fix is to try 10 to 100n in parallel with R23 and if no joy then a few pF in parallel with R20 and then R19.
David makes a good point about possible oscillation due to high impedances. I'd consider reducing the values of R19, R20, R22 & R23 by a common factor of 5 to 25. Many spurious coupling paths are layout dependant so in the worst case, all of the changes may be needed.