After reading much of this thread, I decided to test out the recommended workarounds. I moved my cab block in the grid to the very end, after the time-based effects, and placed the send block right before it, like this:
A/B testing with the cab block before and after the time-based effects gave very little difference in terms of sound. So, what others were saying about the order of the chain is true: the cab block’s position does not really matter in relation to these effect blocks. If you are using stereo reverbs and delays, use a stereo cab/IR. And probably this is also how the Kemper works under the hood: the cab emulation is only added at the very end so that the effects do not need to be run twice in parallel. (Kemper manual link if anyone wants to find it.)
This is a much better approach than splitting a row and duplicating the effect blocks. Yes, it adds one more block (the send), but that should not be a big deal. Try it out and see if you are convinced by the suggestion.
If Neural DPS implemented this request as intended, it would mean reserving DSP and running a whole parallel grid, potentially doubling the resources needed by the original chain. Would this halve the allowed processing power for a preset? At that point, I would prefer them implementing the seamless preset switching… It is just not a good fit for their grid architecture, compared to the linear/rigid signal chain in the Kemper.
Perhaps an alternative would be to have a mixer view for each output and send, where the signal for each is taken from arbitrary points in the chain (like when you do side-chaining). That would make the send and some blend blocks redundant, and would allow for some interesting routing. (Borrowing a bit from Mix several signals to one send-output , Headphone output - Advanced mixer , etc.)