I think that narrativism falls apart when the game world is not real enough.
I find the fantasy world to be a more satisfying construction when it doesn't level in lockstep with the PC's.
I think there are at least two aspects to the "reality" of the gameworld.
One aspect is "causal" or "social/geographic" reality: rivers flow downhill, towns grow up around rivers, people you resuce are grateful, people you are cruel to dislkie you, etc. I think this is fairly important for an RPG that is going to have much depth to its story/theme - it's hard to derive much thematic energy from a setting that doesn't have the minimal sort of coherence with the real world that makes emotional investment and resonance possible.
The second aspect is "mechanical" or "system" reality. In a certain sort of play, this aspect is closely related to the first, either because (i) the mechanics of the game drive the social/causal reality (think Classic Traveller, for example) or (ii) the GM's interpretation and application of causal/social reality is itself part of the mechanical system (this is true of at least parts of Gygaxian AD&D play, I think, and if I understand it properly is also part of LostSoul's 4e "fiction first" hack).
But in my experience it is possible to preserve social/causal reality while divorcing it a fair bit from the mechanics of the game, which take on a bit more of a metagame function. In mathematical terms, for example, the PCs may treble or more in power between Heroic and Paragon tier. But if the gameworld only responds in some much more loose sense - the PCs have gone from local would-be heroes to people of some stature - it doesn't have to hurt the play.
Or to give an example from my current campaign - the PCs start off fighting goblins of mostly 1st to 3rd level, then move on to hobgoblins of mostly 3rd to 6th level, and then gnolls of mostly 5th to 8th level. Those 8th level gnolls could, in mathematical terms, make pretty short work of the 1st level goblins. But in the gameworld, although it is recognised that the gnolls are a bit tougher than the goblins, there is no sense that that toughness is proportionate to the mathematical ratios.
Overall: the maths drives the game in a certain direction, but its interpretation in the gameworld is much looser and more relaxed. As long as nothing happens in the course of the game that forces a direct comparison of the goblin maths to the gnoll maths, no harm done. I've found this to be true even in a pretty hardcore simulationist system (Rolemaster) - none of the players really noticed or complained that the average bandit was around 2nd level when the party was at low levels and around 5th level when the party was at mid levels - it just doesn't come up (in part because no one is paying enough attention, in part because I think it's accepted by at least some players that there will be a degree of opponent-scaling in an RPG). In 4e it's even easier, because there's much less of a sense that the numbers are direct measurements of anything in the gameworld.