In general, I wouldn't set it to the "max" the controller supports. If you try to do the test from xLights type of thing, it will be sending a LOT of extra data thats not needed.
I generally shoot for some happy medium.
My general suggestions:
1) put controllers that wont likely ever change size at the top of the list. The P5 panels, for example, will likely not change much or at all so having them set to the size they need and stick them at the top likely will allow them to never have to be reconfigured.
2) Size the remainder to a bit more than they need, usually enough extra to handle 2-3 extra "average props" at a minimum. In general, I add about a thousand channels which allows moving 3 100pixel props onto it if needed. For me, that's what happens the most. When I go setup in the yard, my carefully laid plans don't always work and I need to move a prop from one controller to another or similar so having a bit of a buffer helps. If the controller is mostly empty (lots of unused ports), I may add even more. If the controller is fairly full, maybe not as much.
3) When doing #2 round each UP to some multiple of 3 so each "first channel" is on a "3+1" boundary (like channel 3001). This just helps keep the "R" of the RGB in the right place. Kind of irrelevant, but helps for testings.
4) Once a year as we get close to show setup, I'll adjust the amount of extra channels for each controller based on estimates for this years show. That's usually enough. If the gaps are big enough, I can move things around a bit and really just need to reconfigure the outputs of that one controller.
THAT said, if you use FPP Connect and have everything setup, reconfiguring the controllers if you have to completely adjust everything takes almost no time at all. I can setup entirely new FPP SD cards in the controllers, make sure the hostnames/ip addresses are what I need, hit FPP connect, and all the controllers are ready to go. The largest amount of time will likely be spent on the "Batch Render" phase to re-render everything.
