How to precisely benchmark impact of subtle settings changes in VPX?

[Markran]

I've recently upgraded to VPX 10.8 release, 64-bit and some new peripherals and it all works well. Since I have an older CPU driving a newer GFX card, medium-ish settings look good but they aren't visually optimal. Maximum-ish settings look great but have occasional stutters on a few 'heavier' tables above 60fps, some tables up to 120fps and many tables above 120fps (my monitor can go to 165fps but my priority is at least 120fps).

So now I need to optimize my settings to find the 'sweet spot' balance between virtually no stutters and the best visual quality my hardware can achieve at 120fps or higher. I already know how to hit F11, play a representative set of games and watch the graph to get a sense of performance. But between VPX and the GFX card drivers there are a couple dozen settings which trade off performance for visual quality. That's a lot of test time per setting across even a half-dozen tables and it's not very consistent run-to-run, especially since stutters get infrequent once approaching more optimal settings, and some settings can change only 3 to 5 fps - and even then, only in certain game situations.

Main Question
I'm looking for a more rigorous and consistent process for testing all these GFX setting permutations that's also, hopefully, automated so it's faster and not so laborious. Is there such a thing?

To be clear, I'm not looking for specific performance optimization advice or what settings work best for you. I'm looking for a better process to identify optimal settings in my unique config than manually exercising tables through enough gameplay modes while simultaneously trying to watch the graphs for blips. For example, even a way to get VPX to output the high, low and avg fps to a file or the command line would help. Especially since the goal is a table of results across different settings, tables and sessions. There must be a better way, right? Ideally, there's something closer to PC game benchmarks which automatically run the game identically and record the performance. Even with the automation it still can take a lot of time but at least it's less subject to random variability and subjective observation. With the current manual approach, even if I put in all the effort, I don't feel like the results are very precise, rigorous or repeatable. I'd appreciate any pointers toward automation or faster, more consistent workflow.