This argument seems like it would benefit from a re-focus.
There are (at least) two broad approaches to the fiction-mechanics relationship in RPGing.
One is to treat the mechanics as a model, in some loose sense, of the fiction, such that the fiction is largely read off the workings of the mechanics. In one version of this, the mechanics are (at least in principle) tightly tuned so that the fiction that gets read off their workings is coherent, verisimilitudinous, etc. The classic exemplars of this are RuneQuest and Role Master.
The second version of this first approach is to set the mechanics more or-less-arbitrarily, or perhaps with an eye on mechanical game balance, and then to suck up whatever fiction results, no matter how absurd. I regard 3E D&D and its variants as exemplars of this.
The second approach is to have a notion of the fiction prior to the mechanics, and to use the mechanics to mediate and perhaps develop this notion. 4e is a clear example: we have a notion that a giant is tough compared to a 6th level PC, and so stat it as a level 6 solo; and we have a notion that the same giant is not tough compared to an epic-tier PC, and so we stat it as a 21st level minion.
Marvel Heroic RP is another example: we don't need the mechanics to tell us that the Hulk is stronger than Aunt May; we know this, and this constrains action declarations and resolutions involving an arm wrestle between the two of them. Only when we're not sure and want to leave the matter open - eg if the Hulk confronts the Thing - do we need to invoke the mechanics.
The second approach is obviously "fiction first" in a way that the first is not: the mechanics can't be invoked or applied, to frame a situation, without some prior conception of how it fits into the fiction (eg what tier is the challenge supposed to be?). This will produce a different play experience from the first approach, in which situations can be framed purely mechanically without forming any view about their relationship to the fiction. What can't be done, using the second approach, is to leave it an open question, to be discovered in play, whether or not a framed situation is relevant in some or other fashion to the PCs. Decisions of that sort have to be made as part of the mechanical framing process.
The idea that one approach is more meritorious than the other, or that the second approach produces shallow fiction, seems obviously false.