There are two problems here:

1. Clearly, the mrml files do not get loaded by the same code depending on what place you load them from. We should have only one way to load a scene and it should be accessed from all the interfaces that offer to do this.
2. How is it possible, that only one sceneview is missing?

BTW: in both cases I started slicer, loaded the scene and took the snapshot. I did not use close scene.

Possible clue: I believe AddData is importing into the existing scene while Load Scene closes the old scene first.

As noted above, I exited and restarted slicer before loading the scene in both cases.

If I don't hide the scene, it appears that the last scene view node is not inside the top level hierarchy node

You should see the same behaviour using the Welcome module Load Data button (or File -> Add Data) as you would see using File -> Import Scene. File -> Load Scene closes the current scene first and then calls Import, and since we were having some scene close issues, fixing those might have resolved this problem slightly.

Looking at the data in question, the mrml file shows the VirtualProbe storage node (vtkMRMLSceneViewStorageNode12) inside of a scene view (Master Scene View) as well as being it's own top level scene view with it's own storage node with the same id. I don't think that storing a scene is supposed to store scene view nodes, I think I fixed a related bug a while back, but the mrml file ADemo.mrml isn't showing a revision number.

svn 19603 fixes import, restore, close crash

Looking more closely at the ADemo.mrml file, there are hierarchy nodes that are missing the pointers to their associated nodes. Nothing's pointing to the VirutalProbe scene view, so it's not showing up in the Scene Views tree on Scene Load. It shows up in the list of scene views available to restore from the tool bar along the top because that just scans the scenes for scene view nodes and doesn't need the hierarchy nodes.

I could write some code that after scene import is done, it makes sure that scene view nodes all have hierarchy nodes, but I'd rather that the mrml scene file was updated to be correct (attached to this bug).