TL;DR flush-fit, lugs, long mounting screws, torque, Loctite == maximum reliability. Especially, if you also rack you’re slide with your optic.
First noticed this with my M18 and Holosun without mounting plate. After about ~3k rounds noticed accuracy drop and subsequently, zero was inconsistent.
Because I’m
absolutely ridiculous I decided to experiment up to seeing which has the most prevention (yeah, I do

like that! no regrets!).
After cleaning the surface and screws, fresh Loctite and torque, adding witness marks, screws start walking again at just over 2000rds. I didn’t necessarily think mounting plates were a scam but I questioned the impact of them - for my own eyes.
Then got an additional plate and grinded down the lugs but flush fit, screws walking at ~1800. Big reduction but already weakened screws.
Did this on an aftermarket slide and pretty much the same results, where not being flush fit and having lugs causes screws to walk faster and increases thread wear.
Obviously, there too many variables and scenarios to control for.
At least on my defense pistols, optics are either flush fit or use a mounting plate/shims and have lugs.
But I do find it a bit crazy that something like an RMR footprint isn’t optimized for and I only have 1.
They should at least including shims for the most popular optics.