As a bit of a PSA, and because most folks probably scan my blog through a reader and don't bother to read comments, I'm going to repost something Matt Finch said the other day in response to the "megadungeon monotony" post.
There are tons of house rules, class variants, new monsters, etc, littering OSR blogs (two thumbs pointing to this guy - just as guilty). Tons of resources on building a sandbox, too. But keeping a sandbox game going, keeping it interesting? Yep, not so much - that's why I thought a note with solid practical advice on *running* the game is worth a special shout.
It is absolutely key that a megadungeon has distinct areas -- and I mean distinct enough that the players can make the distinction based on play, not that the DM happens to know that there's a greater chance of giant rats there. Not only that, but the distinctions need to be meaningful. This allows players the ability to make a decision about where to go in the dungeon. Without that, it is a fungible, undifferentiated, who-cares landscape.
It's incumbent on the DM to either create a dungeon where future missions can be perceived and planned, or to provide some missions the players might choose. Blind exploration gets old if not leavened with some purposeful expeditions that have planning behind them.
Upshot of both those points is that players have to be given meaningful choices and the ability to pre-plan at least some of their activity.
The major potential failing of a megadungeon is not to realize that pre-planning and preparation are fun and integral to the player experience, and a megadungeon risks removing this aspect of the game if it doesn't provide information. Information flow and discoveries are critical components of a megadungeon campaign, or else there is no meaning behind player choices. Might as well flip a coin about our equipment, spell, and direction choices? Not fun.