I keep putting off actually building gauntlet dungeons. I've been thinking a little about why that is. What are the hard problems here that I've quietly, subconsciously hemming and hawing over for like two years now?
One troublesome thing is that I feel like if I'm going to build levels that test the presence and effective use of certain classes, I should keep those levels sort of "balanced" between each other, so that certain classes don't feel like they're useless. This almost implies a horizontal dungeon structure - if class-check levels are in series with descending dungeon level, then some class is still going to feel shafted or "dependent" on another class to get them down to where they shine. And this also dilutes the viability of weird party comps which I was hoping for from class-focused levels.
I'm not really sure how I feel about building "wide" dungeon levels with sublevels testing individual classes. Maybe this is an overreaction to being burnt on Rathell, the widest single level I've ever run. I'm not sure how much patience / tolerance players will have for a single level of difficulty, even if it presents somewhat varied challenges. On the other hand, it might make finding a path down more special (or more intimidating).
One solution to these two concerns / conflicting requirements is thinking more finely-grained about class-check levels and mixing class checks within single, smaller levels. Rather than making a level (for example) a thief check or a cleric check, throw in individual elements that are cleric or thief checks. A level might have a bunch of locks and a bunch of shriekers, testing thief for open locks and cleric for silence, or it might have a bunch of traps and undead, checking thief for trapfinding and cleric for turning (or cleric for find traps...). So thinking about classes as bundles of features that are individually tested, and making testing multiple classes within a single level the standard.
This sounds pretty easy for thief, MU, and cleric - thief has a bunch of skills (which "come online" at varying rates), and MU and cleric have huge banks of spells to test individually. Fighter is the tough one. Clerics can pass AC checks just as well as the fighter, thieves can pass magic sword checks just as well as the fighter. Cleaving checks (if one's game even has cleaves) are often passable by MUs with sleep, fireball, etc. Max HP is a strange thing to test; I guess you could kinda do the 5e death spell thing, conditioning effects on max HP? Rejigger sleep and cloudkill to work off of max HP rather than HD? That's actually a really funny thought. Conditioning effects on a 4e-style "bloodied" status when people are at half, so having bigger HP pools changes when those effects kick in?
I suppose one thing fighters are likely to have that most other characters will neglect (or at least not be raising through ability score tradeoffs) is just brute Strength. Bring out the stuck doors, the portcullises that need lifted, the columns that need toppled, the stones that need carried, the chains that need pulled or broken. This idea is growing on me - what sort of funhouse dungeon is complete without some strongman competition events? So your MU has ogre power, fine, it's only three turns, Fafhrd here can flex those thews all day long.
Another angle on classes, if one were to build multiple smaller (say) 3-level dungeons, would be to tightly-class-theme individual dungeons. Then you're not blocked on getting to the Thief Level by having to pass the Cleric Level. And it's a pleasing idea from the in-world perspective; the legendary thief is gonna build a three-level gauntlet of traps and locks and sheer walls to make sure that whoever gets his gold is a worthy successor, you know? The ruined temple of the Krolm is a gauntlet of strength and hit points and cleaving, it will take a lot of lore and prep to get into the sanctum of the Old Archmagos, etc. The undead angle on clerics is a bit odd; maybe it's what is being kept in rather than out. Or holy places with features intended to keep the unfaithful out have fallen to darkness.
Returning to "test multiple class features within a single megadungeon level", though, the other difficulty I have is in the very first level of a gauntlet megadungeon, for completely new players. You just know their party composition is going to get messed up by casualties and you might not want them to rely on hirelings to patch it up. I would almost be tempted to put "big scores" behind checks on individual class "core feature" like sleep and turning - getting a cleric or MU to the right place and recognizing that that's the right time (eg, to not have blown your sleep on a piddly random encounter) is a good start and a behavior we want to reinforce. Other than that, maybe fundamentals of practice are the thing to focus on, like basic mapping? A caller check, around "can the caller successfully interrupt impulsive players about to do something stupid?" Maybe checks on mundane gear are also a good theme; most of that stuff is well within the purchasing power of first-level characters and mundane gear checks aren't dependent on having any particular class still alive.
Maybe that's an interesting solution to the first-level problem. For seasoned veterans, a dungeon level with few monsters focused on mundane gear "puzzles" could be a bit dull. But for people who have never played before, just successfully executing on the basics of dungeon exploration and random encounters could be a stimulating challenge even without having to worry about tackling serious (eg humanoid) lairs.
So then maybe the second level introduces humanoid faction play and gets more serious about mapping, logistics, traps... What would be a good "capstone" for a three-level 'tutorial' gauntlet dungeon, I wonder?
