@FullWings
A simple example where the sim might not have the required fidelity in the all-engines out case, and the existing usage of the sim would not test the case.
Suppose there is a surface which is hydraulically powered even in the event of single engine failure (it has 2 PCUs, perhaps, one powered ultimately by each engine). For all the "normal" failure cases for which the sim is qualified, all it needs to represent is the reduced speed (and maybe max deflection) of a surface with half the PCUs. During, as you say, startup the surface is totally unpowered - and the aircraft is stationary, so the surface can just "sit there" and no-one is going to care.
But suppose that when that same surface is unpowered, with NO operative PCUs, it then "floats" to the aerodynamic neutral position. that's not something it does in any of the "normal" cases above. So unless someone specifically modelled this behaviour, the sim won't do it - the surface will just sit at the normal/stowed position.
So when you try your all engines out approach in the sim, all the spoilers (for example) stay nicely deployed. the day you do it in the aircraft, half of them float up and significantly increase your drag, and maybe you don't actually make the runway.
The underlying issue is that the sim is not a first principle physics based model of the workings of the aircraft and all its components. Some of it IS physics based (like the equations of motion that actually "move" the flight model) but large chunks of it are empirical, or table/data based (like the actual aerodynamics, which are just data tables for KNOWN conditions, or the systems behaviour, which in some cases is just "effects based"). So the sim cannot reliably extrapolate to other conditions, as the basis of the modelling is simply not set up that way.