I've done each test twice for each setting, so a total of 18+ tests and here's what I'm getting as an average across all of those:
LOD Blending
| Ultra | High | Off |
| Average | 99.5 | 99.6 | 100.1 |
| Minimum | 14.8 | 81.5 | 43.9 |
| Maximum | 166.9 | 129.9 | 125.2 |
Terrain Work
| High | Medium | Low |
| Average | 126.9 | 127 | 127.7 |
| Minimum | 65.9 | 57.6 | 60.6 |
| Maximum | 164.4 | 166.6 | 157.7 |
As you can see there is virtually no difference as I'd consider that within the margin of error. I am noticing the pop-in issue, specifically with LOD Blending, but that just doesn't seem to result in any real/big differences. I'm not sure if it's that my rig is just so strong it doesn't matter, but I'm running with all other settings at max, including supersampling at 2.0 to push things as hard as possible. Oh and I'm exiting to desktop between each test to make sure it's not something in the engine not loading the settings in real time.
Given, that your system does run fine and does not have any issues, the variation in your results shows the render pipeline stalls on some occasions. Checking benchmarks of Planet Zoo, which also uses Cobra Engine, it could be assumed ED is also suffering from draw call overhead on the main thread. Moving the engine to DX12 or Vulkan could solve this bottleneck.
There's might be of a logical problem with the core comparison though. I suspect that the percentage impacts of each of the settings might change depending on the card you're using. You've got the best card you can have right now, which is good, but at the same time, the majority of people who who are going to need to even bother with any of that are people running lower hardware... so the percentage differences might not apply out of the gate.
But this is not as prominent as you think it is. Since the FPS output is also affected by the rest of the system (hard- and software), I recon it is a bigger factor than the difference between each different cards. Additionally the performance difference can be dependend of the content, that needs to be rendered. I hope you can see where this leads.
The only important point from my perspective is, that the amount of available memory (system and GPU) can very noticeably affect performance, if overloaded by the game.
Edit:
@Exigeous: Comment on the video.
Regarding VR performance metrics: Instead of benchmarking FPS in "2D", which can have a very different result over VR, you could have measured the frametimes of CPU and GPU.
Recommended settings: FXAA might look better on static shots, SMAA does a better job on moving images. Especially for VR SMAA gives a slightly clearer picture.
Bloom: Most noteably the red cockpits lights an some ships if the res of bloom has been lowered.
Anisotropic filtering: Your explanation is slightly wrong or confusing. While AA causes some amount (depending of technique and implementation) of blur (by its nature of reducing contrasts on edges), AF actually reduces blur on angled textures.
I assume it will return shortly.
