Yesterday Erik Tenkar, who suffers mightily as my GM, posted about sandboxes and drop-in gaming. Before you read this, go read his post here:
My Sandbox Failure - Not a Good Fit For a "Once a Month Drop in / Drop Out" Campaign
I threw in a comment there, but I'll expand a bit here.
One reason I think a notionally pure "sandbox" doesn't work so well in a drop-in game is that you need time to interact with the environment. The players must be able to engage with the environment regularly, with a cumulative impact, and engage on a larger scale.
It doesn't work so well if you're exploring the Eastern Wastes or taming a small area of Krail's Folly but can't make it to the next game. It's hard to do something that is multiple sessions worth of work if you're playing once a month, nevermind if you miss a month here or there due to scheduling conflicts.
Well, it can, in that Gygaxian "STRICT TIME RECORDS MUST BE KEPT" kind of game with everyone on different time lines, but seriously, tracking that is like project management. If you want a simpler approach, it's easier if time just flows along and missing a session means missing a chance to do something.
A sandbox can work, if it's the right size.
A megadungeon is the right kind of environment. Well, so is any other small-ish, bounded or limited sandbox. It can be a big dungeon, like Jeff Rient's Wessex game from a couple years back.
One reason a big dungeon works well is the whole cumulative and immediate nature of exploration. There is stuff to do right now, so no sandbox project is bigger than a single session on its own. But they can all add up. A large change in the play area in a sandbox can mess your plans up -"Sorry, Mirado and Rul cleared that hex you started clearing while you were out with the flu last month." "Oh . . . I'll start a new project I guess." A large change in a megadungeon is typical - "Sorry, Mirado and Rul cleared that level out last week." "Okay, I head to the stairs to level 3."
I'd also recommend starting and stopping at a convenient end point. I use town and enforce the need to return to town before the session ends. This allows swapping PCs by the same player or adding or removing players without any disruption or gonzo "suddenly, a wizard beams in and the dwarf and gnome disappear!" stuff. You can easily make the PCs all part of a loose grouping and easily explain a new PC in a way that a wilderness sandbox makes trickier.
It's also easier to restock plausibly without messing up the PC's plans. It's also easier to let one-time trips go in without essentially changing the sandbox in some way that throws off the folks not devoting steady play time to the game.
In short, I'd say for a drop-in game, a megadungeon is really useful if you want a sandboxy experience.
But what about rails? Can't I just plunk a dungeon of the week in front of the players? Yes, absolutely you can. Train to Adventure, next stop, Dungeon of the Week! This is our final stop, everyone please vacate the train. It's totally fine. We did that for three sessions in Erik Tenkar's game, and it was really fun. It just means more prep session to session for the GM than doing one big megadungeon. My original plan for my current DF game was dungeon-of-the-week, but a megadungeon turned out to be so much easier . . .
(PS - we'll be exploring a megadungeon soon in Erik's game . . . )