I went through the rest of my algorithms and I'm seeing very inconsistent treatment of these dual inputs. It would be great to hear from someone if any of these are unexpected and bugs, or "just the way things work". It seems the modulation and reverb algorithms are the ones with the worst bleed.
It may be that these are intended to be mono->stereo, but then maybe we could have a setting to force dual-mono?
To test this, I ran the A/B setup above, enabled Input 1 (Amp A), put Amp A on standby and listened only to Amp B (Output 2). The H9 is running in KillDry=Off (i.e., as a stomp box). Where available, I put the A/B Mix settings on A10+B10 and Wet/Dry mix at 50%.
1. ModEchoVerb is the worst offender, the 100% DRY Input 1 signal joins the effected signal on Amp B
2. The following algorithms spilled 100% of the wet Input 1 signal into Amp B
3. The following algorithms had a minor spill of the Input 1 wet signal into Amp B
4. The following algorithms were clean, but you could hear modulation artifacts in Amp B
5. The following algorithms worked as I expected (no signal on Amp B)
DualVerb (A/B mix set to dual mono)
#1 and #2 above seem like bugs to me, and make the pedal unusable in my A/B configuration (my clean guitar sound will get distorted reverb, not good).
#3 and #4 are OK for rock-and-roll (the enabled amp will cover the slight bleed) but it was noticeable in the standby scenario.
#5 were the only two algorithms that seemed to actually keep the outputs separate.