We never claim that the switch from app view to multitasking view is the reverse of launching an app, we just say that the *animation* is the reverse (zooming in and zooming out).
you seem to imply that transitions should be limited to 2 views
Nope, definitely not. There can be any number of views, they just need to be connected by animations that make sense when you put them all together.
But clearly that would be less productive, requiring more steps to get to where you want to be.
As the Mac OSX example shows, that doesn’t have to be the case: It’s the same interaction, except instead of zooming, the app just scales and moves horizontally. That wouldn’t be any more complex than the current animation, but a lot less confusing.
As I pointed out in my response to Hans Christian above, employing semantic animation doesn’t mean you can’t have complex interactions or spatial structures. It just forces you to think about the bigger picture, how everything works together. In my experience this constraint is very useful, because if a design doesn’t play well with semantic animation it’s often because it’s more complicated than it needed to be in the first place.