Nagol, I've got not doubt that the group has to be on the same page, more-or-less.
In my own group there's one play who's more interested in 4e as a gamist excercise than a thematic exercise, but happily the 4e mechanics (or, at least, the way we use them) mean that that player's pursuit of tactical victory neither impedes nor is impeded by the other players treating the tactical game as a vehicle for expressing theme. So I think a bit of accommodation can sometimes be possible.
There are two glosses that I would want to add to your post, though.
First, the idea of "GM intervention" I think is a little misleading. Very few games have a "justice system resolution table", so the conequences of a murderer being "turned in" are always up to the GM to determine. There is nothing distinctive about narrativist play in this respect - what is distinctive is the considerations that inform the GM's determination.
I've seen them, but they aren't necessary most of the time. Although the result is defined by DM fiat, there is only so far the DM can walk before damaging suspension of disbelief, establishing precedent for PC immunity, and/or extending the punishment of justice to the innocent players. All of which negatively affect the campaign and play group. After all, long-term incarceration is often more damaging to the PC than death in D&D. So if the DM is trying to keep the PC in play despite the social consequence of a capital crime, he is limited to consistently less likely legal results of not guity verdicts, wereguild, exile, or suicide missions.
Not guilty verdicts and wereguild generate a feeling of immunity as the PCs expect consistency of ruling and tend to fall into riches during play. Exile and suicide missions force the other players in the position of accepting the same punishment or condemn the rogue player themselves. This can poison camraderie more than the DM providing a response since the DM is already an outsider to that team spirit.
Second, I don't think that narrativist players are particularly prone to being disruptive or derailing games, or pose any extraordinary problem in terms of expectation management. Hardcore gamist players, for example, who focus on PC optimisation and who approach every ingame situation from a fiction-light pawn stance, are a well known issue for simulationist GURPS, HERO, 3E or even AD&D 2nd ed play.
And a certain sort of simulationist player, who wants to roleplay out every shopping expedition , for example, and who plays the character to the exclusion of engaging the situations the GM is presenting to the group, is disruptive of mainstream D&D play focused on simulation as a chassis for gamism, because this sort of simulationist player refuses to "step on up".
I'm unsure that narrativist are more innocent of disruption. I have two players at my table that try to explore narrative themes. One pursues it without group agreement and derails sessions, group goals, and generates a fair amount of conflict within the group. It's at least as disruptive to the group as a the fiction-light camp -- the group has less internal strife with them even though the consequences of their actions are at least as damaging to the group efforts because it isn't pursued as a solitary choice.
So I agree that some sort of mutuality in the group is important, but I don't think that simulationism provides any sort of privileged safe harbour. In fact, in my (admittedly limited) experience it is fairly easy for a narrativist player to drift a group in that direction - you keep following the GM's plot hooks, but treat the GM's storyline as simply a backdrop against which thematically driven PC-to-PC interaction takes place, which interactions in turn inform the way the group engages with the GM's storyline. This is perhaps not the most functional RPGing of all time (it depends a fair bit on how the GM responds).
Simulationism isn't a safe harbour, but it can provide the framework of shared expectation that allows the players to continue to engage in the campaign in a personally meaningful way. More important is consistency of application and response from the DM. It isn't simulationism to say "Your character is accused by his compatriots of killing the respected necromancer X. As he is considered both extremely dangerous and a flight risk, he's being held without bail for the investigation to complete. Based upon your character's statements and witness statements, a guilty verdict is likely. Would you like to roll a replacement character or wait for the trial?" It is an application of narrative power that helps maintain the imaginary society at a predictable and understandable level for the group and provides understandable and predictable consequence to meaningful player choice.
But at least in my experience, it is evidence against the notion of simulationism (or very exploration heavy gamism of the classic D&D variety) as a default approach to RPGing.
Exploration-heavy gaming is by no means the default for RPGing -- take for example most superhero games where the characters are a reactive and defensive force. It was the stated default for D&D though that became the unstated default and then supplanted by the unstated default of heroic action in a developed plot.