From what I’ve understood so far, Blender’s compositor is limited to the render output resolution. If the output resolution is smaller than an image you want to composite it with, the compositor will restrict its output to the render resolution. It makes sense to me why it does this because it doesn’t have any more scene data than the render resolution to composite with.
marmor0 is using render region to save render time as explained by the numbers in the first post. It seems like a bug that a render region in one scene is disabled when you are rendering two scenes. Probably just not something anyone thought of doing.
I think the only real solution to this with Blender is like you said, adjust the output resolution and focal length to match the render region they have setup now. This should have the same performance savings as the render region. Then they can render a second empty scene with 1 sample(something I forgot to do in my blend) to do the final composite at the final resolution.
edit: @marmor0 I found this addon that should let you convert your render region into your camera settings: