Are you using ERG mode when TR is controlling the trainer? I think Zwift will be using slope mode (at least not ERG) when just riding. I would hope that power reporting accuracy would be the same for the kickr in both modes, but this is a difference. I don’t think this part has changed from your before/after tests, and if not this wouldn’t explain your discrepancies.
I could see how the kicker ERG mode power smoothing could cause a difference here, but from what you have described nothing has changed from your before/after tests other than the bike. I don’t have an explanation of why power smoothing would affect 1 bike and not the other, but you certainly want it off if you are comparing power. (I guess gearing could have an effect, but I would expect large gearing changes to be required to have a significant effect.)
I agree that swapping protocols is worth a try, and I think with TR alone trying both ERG and slope mode as well, this would help tell if the trainer resistance mode is a factor or not.