As to this CON: Portrait mode shallow DOF works in real-time on screen, so why not in video mode?
I've had some shots where it doesn't even sort out an image on screen or in a still. For instance, if there's not enough distance between your subject and a semi transparent background like a chainlink fence. I've seen portrait mode get confused and blur the fence out in some areas and not others. Same if the subject is in front of a window that the camera can see through in some spots, but where the deeper background is partially obscured by reflections. I would expect trying to apply that level of ai to a similarly confusing situation in video would create strange and awful results.