With regard to 3D indoor scenes, performance is entirely related to the complexity of the scenes & how much effort is put into loading/unloading non-visible areas etc.
For a couple of data points, here is the official Godot third person shooter demo:
Various people have experimented with voxel worlds and other things which push the limits of the engine & are viewable on YT.
In short, with 3.0, Godot isn't competing with, say, Unreal in the 3D realism stakes but it still has more 3D capability than most people who start to make a 3D game will actually hit when the scope of the work required becomes apparent. :)
Any game where camera frustum culling can easily cull the number of object rendered. So when the camera is pointing in a fixed direction and you have control what moves in front of it.
That would only leave 3D models + minimal scene (e.g. top-down/isometric games), which minimizes the amount of occlusion?