The only reason I can think of to make an interface that mimics a physical object is that the new interface is more likely to be instantly recognizable and intuitive to a user looking at it for the first time. In this case you're designing the interface to work best for a new user, which generally means it won't work as well for an advanced user. I would guess that designers of basic calculator apps, like the example in the post, are designing for the new or occasional user and assume that anyone doing lots of calculations will move on to something more efficient.

The problem with extending physical mimicry to more complex software is that the interface ends up adopting design compromises made for the physical world. I remember the idea of a "virtual shopping mall" where you would virtually walk through the mall, look around and pick a store to go into. But a shopping mall has to be a certain size because it has to hold a certain amount of stuff. It takes time to walk from one end of a mall to the other, but there's no reason to extend that compromise (made to gain physical space) should ever be made in software. For anything beyond a very simple application you're better off asking the user to learn a few things about how the software works and designing the interface for its medium.