Actually, Tolkien's Dwarves-Jews analogy in that very interview is merely philological in nature: "The dwarves of course are quite obviously - wouldn't you say that in many ways they remind you of the Jews? Their words are Semitic obviously, constructed to be Semitic." (See
here for a transscript of the interview). In one of his letters, Tolkien already presented the comparison with precisely the arguments davem mentioned.
But that's not the topic.
The problem with 'canon' is as far as I can see not just simply to be reduced to a single one. Rather, we have different grey and black spots in regard to the Legendarium.
Fordim's initial examples show this -- the origin and nature of orcs was at times uncertain, but quite clear later in the development of Middle-earth. However, the man still changed his mind on it several times. I believe that such a case provides us with the same problem as all texts which exist in diverging versions without a definite authority or definite solution to the complexity. Older mediaevalist science, for example, tried to construct an 'author's text' from the extant material, looking for sensible compromises and judging by their own ideas of taste and style. The result was of course a stab in the dark in regard to authencity. The more modern point of view is to take the conflicting versions for what they are, and rather ask why it says so in one copy and differently in another.
It appears to me that most discussions of such matters in the Legendarium however follow the old way, for better or worse. Nevertheless, while it is assumed that differences in medieaval
Minnesang for example are there for a reason and presumably often intented by an author adapting his work to his audience, Tolkien discussion has a mixed blessing of its own: we can safely assume that the Professor had quite clear ideas about the 'truth' in his stories.
When asked about a matter such as Balrog wings, it is in fact most likely that he could give a simple and precise answer. Having that in mind, discussions of such grey areas are either confined to guessing Tolkien's mind or being content with little, none, or conflicting evidence.
I would like to think that all of that doesn't really have anything to do with
interpretation, which is arguably wholly inadeqaute at answering questions within the Legendarium. Condemning Gondor's autocratic ruler from a modern point of view does not yield anything useful about the facts of the monarchy in the Legendarium. Calling the oligarchic Ruffianism a worker's class revolution is a perception noone in Middle-earth would likely have had either.
That being said, I do believe that there are very definite and easily recognizable boundaries of the canon. I see no point whatsoever, in any form of literary research, to question a 'fact' Tolkien gave us, within or without the fiction, since anything relating to the Legendarium is necessarily part of the fiction. What would be the point of denying the authority of a quote such as the one from the Letters, explaining that Sauron was of human form? One might get a different idea of Sauron when reading the books, and that in itself is interesting, but the fact is part of the whole. If a reader refuses to acknowledge certain parts of a work of fiction (or actual fact, for that matter), there's nothing I can do about it, but that person's position in a discussion forum is difficult to say the least. Conversely, I doubt someone with a clear idea of his own which one is unwilling to give up would ask such a question anyway.
Sub-creation in that context is definitely in accordance with the nature of the Legendarium, as long as the differentiation remains. I can't say much about the RPGs, but the canon rule there serves a very important purpose which is not primarily that of 'staying true to Tolkien', but rather to keep the games sensibly enjoyable.