Re-edited: 04.08.18 10:38 pls re-read.
You expect answers for things that nobody can easily and exactly tell. The majority of customers do not have these kinds of problems.
Whether the Sonnet card gives you advantages over Thunderbolt is not predictable, but you could have tried. Google for "MSI" (message signalled interrupts). I think its a chance as it fixed my problems as well which arose when I added on top of the 2 UFX+ and ADI-2 Pro too many additional USB periphery like Bluetooth adapter, USB3 hub. So I think under certain conditions it still makes sense to fully isolate recording interfaces behind a dedicated controller (be it USB or Firewire).
Strange is the thunderbolt didn't solve the issue, which makes me think whether other surrounding conditions either cause or simply add to the problem (as you said it became better, but was not gone entirely). Somehow it seems to be load related as you have no issues with smaller projects.
You are recording at 96 kHz which definitively puts more stress to a machine (CPU, interrupt and i/o related).
Maybe your system is overloaded for the things that you intend to do, be it CPU or whatever.
I told you try Cubase and to freeze VSTis, but you didn't check. And if 96 kHz is not possible I would simply pragmatically use 44.1 kHz, which in most cases is fully sufficient. I doubt that you will hear any differences. People use 88.2/96 mostly for recording of classical music.
The best might be, that you consult a PC specialist for Audio Workstations like Xi Machines in Germany, they will surely also sell to outside of Germany. They can give you then consultancy what you require and then I am pretty sure you can enjoy every RME recording interface. Maybe a better / cleaner approach than playing trial end error for too long.
The machine simply need to have a good and proven design (i.e. in terms of mainboard / BIOS) and components, then also enough CPU power for the type of projects that you use. You need to discuss with them whats more required for your type of project. Single thread performance or the number of cores. This might also be depend on the DAW that you use.
I personally would recommend to check out Cubase Pro, as there you can freeze projects easily (but still changing gain, panning, etc) which is a simple but very important possibility to reduce CPU load in any project.
Maybe you are also doing mistakes in your workflow, that you mix with already putting mastering plugins into the project. Maybe a strict separation of Mixing, Downmix and then mastering only with one stereo track - which is the recommended workflow anyway- would also help.
I had really good luck with my system as I got a tip in a recording forum to take Xeon based systems with Supermicro mainboards (being used in the enterprise sector where you need to have reliability to be able to hold your SLAs as an integrator and vendor). Many builds of Xi-Machines seem to be based on Supermicro boards and Enterprise components.
BR Ramses - UFX III, 12Mic, XTC, ADI-2 Pro FS R BE, RayDAT, X10SRi-F, E5-1680v4, Win10Pro22H2, Cub14