My recommendation is to try both configurations and see which one works better.
I'm likely going to work off a clean 3.1 install. How do I go about testing various configs, configuring gain, etc. without the mutability/jprochazka performance graphs?
My (repeated) recommendation -- figuring out what helps/hurts is difficult as traffic changes day by day. Because traffic changes day to day, you need a reference to compare to, a reference with performance that also changes with those daily traffic changes.
(1) On your stats page, look at Nearby ADS-B Sites. Pick one close to you that has similar numbers and has been around for a while. This is going to be your reference.
(2) record your numbers and that site's numbers daily to see how they track. The idea is to find a site that tracks your numbers, not in values necessarily, but in the daily ups and downs. I record things at around 0030Z, or when I get home.
(3) when it looks like you've got a good reference, start making changes! Make a change (I tend to make changes late at night when traffic counts are low, but still there so I can be sure I didn't completely bork things). Record your numbers for a few days to get an idea what that change does.
(4) repeat -- make a change, watch the numbers, and learn.
(5) tell us what works and what doesn't!
Fireside story on performance monitoring--
Once upon a time, many, many moons (decades!) ago, I worked in the mainframe computer racket doing system software -- process scheduling, I/O management, file systems, that kind of thing. Our system (CP-V running on SDS Sigma 5/6/7/9 systems) also had a lot of built-in performance monitoring and tuning features. This was in the days when the OS was written in assembly code and computers remembered using little donuts made of rust -- before the days of modern DRAM.
Occasionally we'd get really enthusiastic (nay, rabid) system folks at customer sites who tried to tune the last little bit of performance out of their systems and would come to us (and me) with wild performance claims and questions.
If they indeed sounded a bit unhinged, we asked them to perform their set of tests varying a particular parameter, SL:BB, and let us know what effect this parameter had on performance. What value of SL:BB gives the best performance? The worst? Run your performance tests varying SL:BB and get back to us.
The correct response was that SL:BB doesn't do jack for performance. We'd planned to do things with that parameter, and the ability to change it and report it was built into the performance tools, but down in the guts of the operating system, it wasn't connected to anything.
Yet we had folks who were adamant that a particular value of SL:BB tanked performance, or really improved performance, or that performance was all over the map as they changed SL:BB...
It was a most valuable tool...
End of fireside story.