Joe Browning made an interesting post here on the topic of dungeon design, and I posted a response -- which was promptly gobbled up by the internet and disappeared.
What I wanted to say was actually a bit off topic from Joe's post in any case. His point was that as a matter of realism, dungeons would be built without the sort of multiple-entry, circular-pathway, multiple-connection design that is the "conventional wisdom" of megadungeon design. The only real part of my lost response that I'll rewrite is the point that designing a good tactical environment for gaming trumps the desire for realism in designing the area. I'm not dismissing Joe's point, I just think that there's another layer in addition to what he points out, and that I think the second layer is actually the more important one.
But the second point I made, and the one that I'd like to develop somewhat more (ie, what I'm willing to re-type from the lost post) is that most of the canon wisdom Joe points out is addressed to designing megadungeons -- and does not apply to adventures that take place in a smaller area with a defined mission. Megadungeon canon is specific to maintaining many possible routes and presenting the players with a multiplicity of options to go through a large area.
Lairs are entirely different, and indeed many of the things that are specifically "good" for megadungeon design will cause problems in a shorter and more defined adventure. Multiple pathways? Better keep the number limited, or the party may get lost and sidetracked through no real fault of their own. The potential for bogging down increases -- there is an optimal number of pathways through a lair that is far, far smaller than what you'd want in a megadungeon. Lairs needn't necessarily be linear, but the design of the tactical space needs to be focused upon the adventure's objectives. For a megadungeon, that's exploration, but for a lair it is accomplishing a specific objective that's (usually) been set in the adventure rather than by the players.
For a lair, I would say, the different tactical pathways should contain some sort of circular route for potential backtracking, but clear dead ends are just as valid.
Also, the backstory of a lair is infinitely more important than it is in a megadungeon. One of the things a short adventure should do is to clear up pretty much all of the mysteries (except, perhaps, a mystery that creates the bridge to the next potential adventure). Lairs have a stronger emphasis on fact-discovery than on terrain-discovery. Lairs can certainly be straightforward without a COMPLICATED backstory, but if the party defeats the monsters and finishes the area, they should pretty much understand what was going on there.
Thus, in terms of the structure of the map, it can be seen not just as a tactical map but also as the floor-plan in which facts or clues are located. Possibly they are scattered, or possibly there's just one big final revelation if you can get to the "last room."
I don't think that I can really resolve these thought into any set of rules, but it's worth thinking about the fact that whatever "rules" there might be for designing a small dungeon or a lair would be considerably different from the design principles used in creating a dungeon that's intended for unlimited repeat visits.