I think the other point the author makes is that it's not possible to write efficient code that is entirely abstracted from the underlying data (see his loop examples).
i.e. if you have to write your code in a specific way to make the ORM behave correctly (constantly thinking about what kind of sql your code is generating), then the abstraction becomes a lot less useful.
i.e. if you have to write your code in a specific way to make the ORM behave correctly (constantly thinking about what kind of sql your code is generating), then the abstraction becomes a lot less useful.