Can one speak of natural expectations where magic is concerned? After all, magic can potentially do anything, or nothing.
The case for the rules as the laws of physics of the gameworld (an idea with which I don't agree btw) is, I think, strongest when it comes to the rules for magic. Because magic doesn't exist in our world, it's uncheckable. With the rules for wounds, dying, normal healing etc we can look at the way these work in our own world and say, "These rules don't make sense. They are not a correct simulation." With magic we can't say that the rules don't properly simulate, unless they are explicitly simulating particular works of fiction, which D&D doesn't do, imo. So our conception of what a cure light wounds spell, or other healing magic, does, comes only from the rules.
It's interesting that the dying words example involves a conflict between magic, where our ideas about how things work come from the rules, and real world, or rather adventure fiction, dying, where our ideas do not come from the rules but from fiction.
That said, I'm not sure if the rules alone inform our ideas about magic. Fiction plays a part too, concepts of power hierarchies such as that a god's or an artefact's magic would be stronger than a wizard's, the way skills are learnt in the real world and so forth.
We might also form principles from reading some of the rules, with which other rules seem to conflict. For example we might think that a spell is too powerful for its level.
You're correct, of course, that magic as physics and the "natural expectations" regarding magic will not be uniform. Let me be clear: I think in the game we can accept that the potential of magic is essentially limitless and that "a wizard did it" is perfectly coherent. That, however, does not cover all natural expectations. These include that the magic potential of any given caster is limited in various ways (level/training, internal magical potential, agreements with patrons, etc.) such that the potential of "magic" is arbitrarily large if not actually infinite, but the potential of a given caster at a given moment in time is finite and subject to boundaries. These boundaries are understood in the fiction of the setting, enforced by the mechanics of the rules, and usually present in the fiction we are inspired by. In short, there are no "actual infinities", at least not for PCs, but these natural expectations in no way require us to think of magic as the physics of a setting.
Now the precise nature of these limits is unimportant, all that matters is that they exist. Because if I can imagine a wound I can heal with one spell but cannot with another, I can always suppose there exist wounds more powerful than what I can heal. I'm not speaking solely of hit points. For example, in a setting where the inhabitants have limbs, it is a natural expectation that these could be removed, magically or otherwise. If the rules do not handle the details of this situation, yet we believe that at some point limb removal will result in death, I can imagine spells that could heal some sorts of limb removal but not others.
Applied to the idea of death, it stands to reason that a possible state in the setting is the mortal wound, in the same way the limb loss is possible. Some types of magic will suffice to heal it, and perhaps eventually there is magic which could suffice to heal any mortal wound imaginable. However, unless all magical healing has this potential, it is quite consistent to say that there exist situations where someone will die, has not died yet, and there is nothing one can do about it. I don't think this relies on an idea of simulation, it merely relies on the fact that magic is limitless in the abstract while our characters are not, at least at any given moment. The specifics will rely on the rules at hand, but if the rules are considered incomplete and/or inaccurate for some situations within the game world, there is always room for these natural expectations to guide the DM's decisions.
Your points about the interaction of the rules with fiction, and their conflict are well taken. In another thread Firelance suggested that the most important thing for fluff-crunch relations is that they match well, regardless of whether or not one of these is considered more fundamental than the other. I agree with this wholeheartedly, and I think the same holds true here. If my thoughts about the natural expectations about magic aren't total crap, I think they offer an important constraint for iterating the rules/setting relationship.
Cheers.