This point was made long ago by J.L. Austin in (I believe) Sense and Sensibilia. Austin points out several things about "real", among them that "real" is substantive-hungry: You can't answer "Is such-and-so real?" without asking first, "Is it a real what?"

A decoy duck is not a real duck, but it is a real decoy -- whereas a rubber duck is not a real decoy; and a decoy coot might be mistaken for a decoy duck if you know little of waterfowl, but isn't a real decoy duck.

There is no sense of "real" that applies to all substantives that we would describe as real. The word makes sense only in contrast to specific ways of being unreal: being a forgery, a toy, an hallucination, a fictional character, an exaggeration, a case of mistaken identity, a doctored picture, etc. It is these negative concepts, and not the concept of "real", that actually do all the explanatory work. "Real" is both ambiguous and negative.

Err no! He says that 'real' means something like causally accessible from where we are. It's something like "from my perspective I am real, but from the perspective of a fictional-me in a fictional-universe, I am not, while the fictional me is real". Except this is not a very helpful way to define 'real'. There is no meta-realness, but relativistic-realness is quite as useless. Drescher dissolves the issue, by reducing 'real' to something like "whatever we can possibly get at from where we are in this universe".

It's hard to define 'real'; it's not clear that it's doing any work.

"Hard to define" and "not clearly doing any work" are distinct properties; I'd agree about the former and not the latter. I do find it difficult to give a definition of "real" that isn't going to break when dealing with unusual border cases; but nevertheless, if I consider the question "Is Harry Potter real?", or "Is Barack Obama real?", or "Are atoms real?", then the two possible answers I could give for each will imply distinct models of reality that anticipate different experiences, and furthermore the word "real" can transfer such a model into someone else's mind pretty successfully. It doesn't particularly seem to have any of the characteristics of a non-descriptive term.

If "real" is an honorific, then it can also be used as a descriptive term.

If "Gary Drescher is not real" is false, then clearly we mean something by the word, which makes it a bit tricky to show that it's not meaningful. Maybe you could show that real and unreal things have identical properties, aside from their "honorific?" Monsters under the bed refute that one though...

As in my reply to gwern, I was thinking "If "real" is an honorific, then it can also be used as a descriptive term."