For fun, and since I had some time, I reinstalled RamCache III and tried your method of just increasing render scale without lowering LODs.
When testing your method with max LODs, my FPS was in the single-digits to low-teens by the time I could get GPU-bound again. Rolling back LODs to get my frame rate back put my numbers right back to where I had them before I started testing. I realized that there was no way that was going to work for me and my rig, so instead I spent the time to check every single setting, aside from LOD and traffic, one by one, to see which ones affect the main thread latency.
None of them did.
It’s a long, methodical, detailed writeup, but you might find the results interesting, especially concerning the GPU-bound settings:
TL/DR; CPU latency was only affected by LODs and traffic. Nothing else. However, setting traffic from 100 to 0 was enough to allow me to DOUBLE my LOD settings, and bump up my render scale.
I figured if anyone would find that bit of methodical analysis and testing interesting, it would be you.
Let me know what you think!