Again, I am not certain if the schematic I showed above for the PH-1 is identical to the PH-1r, but it likely has one characteristic that might be the problem.
Normally, in many commercial modulation pedals, the "bypass" consists of a single FET that completes or lifts the connection between the modulated signal and the mixer stage. In the case of the PH-1, at least, you can see that the control line from the flip-flop circuit turns Q5 and Q6 on and off simultaneously. What it looks like is that in bypass mode, Q5 and Q6 are off, such that the phase-shift signal does not reach pin 3 of IC2, and R18 (5k6) is placed essentially out of circuit. With the mixer-stage's gain set by R21 (10k) and R19 (1M), it is effectively a unity-gain stage. When Q5 turns on, however, R18 is placed in parallel with R19, raising the gain to roughly x2.79.
I gather the gain change is to compensate for the loss in volume produced by the notches, and maybe even to provide just a tiny bit of increase in overall level for solos. It is easy to imagine that a 5% tolerance on R18/19/20/21 could lead to the gain-compensation in "ON" mode being adequate in theory but too much in practice. For example, if R21 was on the high side (a bit higher than 10k), and R18 was a bit on the low side (lower than 5k6), the gain would be slightly higher than the theoretical x2.79 in effect mode.
Does that make sense? Does it fit what you hear?
NOTE: Although the board has a different layout, much of that can be attributed to use of 4 dual op-amps instead of two quad op-amps. My guess is that there is really not very much changed in the basic design except for the addition of regeneration/feedback.