In my opinion whether derived tables and/or CTEs are "tables in any sense" is not really relevant. What is relevant is that they can, and probably should, be used in many situations where many would use some sort of temporary table. For example, they have a significant scope advantage over the other pretenders. That is why it makes sense to discuss them in an article on temporary tables.

As far as the relational model goes tables, view, CTEs and other relations are supposed to be equivalent anyway. It is only when you look beyond the relational model and consider things like performance that the distinction should become relevant. I am not saying performance issues are not important, but I think they should be discussed in the context of a relational approach.