First of all your opening post was great. I'm actaully giving it a go now, just wish I had a nvme drive hooked up to the PCIe bus.
There's a couple of points regarding Linux kernels.
Sometimes the latest kernel may not be the optimum for your hardware.
The kernel you are using is a "release candidate" whilst usable for all intents is still going through some work.
The 6.1.y branch works fine on the RPi4, I think the kernel in the Linuxcnc image maybe 6.1.54
The RPi platforms with regards to Linuxcnc are a little different to the x86 platforms. Your Rpi4 & Rpi5 hardware is what it is. Whereas you may need a specific kernel dpending which x86 motherboard you buy.
The Rpi branch that appears to be active is the 6.6.y branch, so one could assume this is the branch that is receiving the most "bug fixies".
Simple things like isolating cores 2,3 will see latency drop. There doesn't seem to be as many options/tricks for latency on the RPi platforms as the x86 platforms.
I would chose to use a kernel that is a couple of version behind the the latest and greatest if it more tried in the real world and has better latency.
In my discussions with royka we talked about the fact that a kernel that is a little older is sometimes the better option.
Due to the issues with wayland I would recommend the use of the image form the Download page, Xorg is used. Plus it's easier to deploy
Mind tho I'm not saying this is the only way to do it.
What I would suggest is trying some of the earlier kernels to see how performance differs.
One thing regarding the 6.6.y branch.
The latest version is 6.6.70 which I couldn't find any real time patches for.
The previous commit is 6.6.65 for which there are rt patches, but for some reason the overlay's aren't in the source. Which is a pain when building an image using Rod's image builder.
It seems, ATM, the deeper you dive it can seem a little confusing.
Sorry for the multiple posts but this editor is PITA.