Run repo mission, baretail the rpt log (highlight ZBE if you'd like) and look at FPS. It's automated. Adjust sleep as necessary for your machine's performance (in init.sqf).

"ZBE 144 Groups, 144 Units, 7 NoCacheFPS, 30 CachedFPS, 54 DeletedFPS"

7NocacheFPS means with 144 Groups, 1145 Units. 30CachedFPS means 144 groups 1001 "cached" units. 54DeletedFPS means 144 groups, 144 units.

No repo mission test:

Spawn in 744 units (or any high number) then execute:
{

                _x disableAI "TARGET";
                _x disableAI "AUTOTARGET";
                _x disableAI "MOVE";
                _x disableAI "ANIM";
                _x disableAI "FSM";
                _x enableSimulation false;
                _x allowDamage false;
		_x hideobject true;

} forEach allUnits;

Keep an eye on FPS.

Then do:

{deleteVehicle _x} forEach allUnits;

Notice FPS returns to pre-createvehicle state.

I've tried removing all weapons, stripping all gear, etc thinking it could be something to do with equipment or weapons but no change in FPS. It's also interesting to note that:

_x enablesimulation false;
_x hideobject true;

by itself (without any of the disableAI commands) results in the same or *slightly* lower FPS.

What is still simulating?

One could roll their eyes at such a "performance" issue as most missions don't spawn in so many units (most create/delete during mission runtime) however I argue that createvehicle and deletevehicle sends packets which impacts netcode performance which is multiplied with high playercount.

enablesimulation false objects do NOT send packets and would be much desired IF such a "caching" method had comparable performance to createvehicle/deletevehicle.