Edwards isn't against setting per se, though he does have a "handle with care" approach. A bit more from
the same essay:
However, neither Setting-based Premise nor a complex Setting history necessarily entails metaplot, as I'm using the term anyway. The best example is afforded by Glorantha: an extremely rich setting with history in place not only for the past, but for the future of play. The magical world of Glorantha will be destroyed and reborn into a relatively mundane new existence, because of the Hero Wars. Many key events during the process are fixed, such as the Dragonrise of 1625. Why isn't this metaplot?
Because none of the above represent decisions made by player-characters; they only provide context for them. The players know all about the upcoming events prior to play. The key issue is this: in playing in (say) a Werewolf game following the published metaplot, the players are intended to be ignorant of the changes in the setting, and to encounter them only through play. The more they participate in these changes (e.g. ferrying a crucial message from one NPC to another), the less they provide theme-based resolution to Premise, not more. Whereas in playing HeroQuest, there's no secret: the Hero Wars are here, and the more everyone enjoys and knows the canonical future events, the more they can provide theme through their characters' decisions during those events.
In designing a Setting-heavy Narrativist rules-set, I strongly suggest following the full-disclosure lead of HeroQuest and abandoning the metaplot "revelation" approach immediately.
He elaborates on this sort of approach to setting
here.
I think this is another interesting feature of 4e: the core ideas of the setting are revealed in player-side rather than GM-side material. (Some later books moved away from this, as part of the general drift of 4e back towards a less radical approach to D&D design and publishing.)