Just thought I'd ask. For games in the wake of Mechwarrior, Incoming, and Urban Assault...Just how long should levels take (a competent player) to complete? How about a cooperative game? Keep in mind that this in the context of a very long campaign.

Well, it really depends on what's going on. MechWarrior (3,4) had levels that were around 10-15 minutes long. Some battles took more, some took less.
I don't think there somebody should put a time limit on levels, as long as what's happening on the screen is interesting. The experience is far more important than the time between loading screens. You can achieve this by changing objectives , inserting dynamic events, making a great combat AI, etc. Hey, even a time acceleration feature would work in some cases. Take a look at Operation Flashpoint: 50 levels of pure fun. :)

One thing that is definitely important is your save system. For single player games, if you make long and complicated missions without the possibility of saving (see Rainbow Six) you need to make them relatively compact, short or extremely well built in order to avoid frustration.
If you use a checkpoint system you have to design the levels around it. The best recent example of a checkpoint save game is Far Cry, because the levels weren't linear. Here you can have long levels, but you have to be careful how you place the checkpoints.
For a classic save-anywhere system you can ignore the things above, and just concentrate on making the game entertaining.

Bottom line, time is irrelevant as long as you avoid player frustration.
The same things can be applied for a cooperative game.

Vaipa said it well. The key here is to a) keep things continually interesting for the player, and b) avoid making the player redo lots of content if he screws up. You can solve this by making shorter levels, having frequent (and well-placed) checkpoints, or by simply allowing the player to save the game at any point.

If the user will lose more than 15 minutes of game time by making a mistake (such as getting killed), the user will be frustrated.

In MechWarrior 4, you couldn't save within a mission, so missions should be complete-able in 15 minutes. Sometimes, the user may crawl stealthily and snipe, and take much longer, but that's OK if there is at least one 15-minute path available. There was one or two MechAssault scenarios that were a little long, but they seemed to have fixed that with intermediate checkpoints in MechAssault 2.

I think the checkpoint system in Far Cry worked very well, btw. There was only one stretch in the entire game that didn't have a checkpoint in the middle, where I feel it should; perhaps I just took an un-expected route around.

I would say at least 10 minutes, anything less might annoy the player (especially if levels take a while to load), but not over 30min. Or in other words, long enough to be satisfying, but short enough that you can play a quick game while you're waiting for the pizza guy or something.

Nuvem makes a good point: Loss of game time to death or restart can be a hassle for players, but constant loading of new levels is just as bad.

The Chronicles of Riddick for the XBox suffers from interruptive loading sessions. An otherwise engaging game is chopped up by that little loading thing popping up every third or fourth door I walk through. Vexing. The poster child for this phenomenon is, of course, Riven on CDs. Having to switch discs every time you ride a trolley had me pulling my hair out. I never finished the game, for just this reason.

"How long should my level be?" is sort of a bogus question. Levels were (and still are) a means of dividing levels both for closure (so you have a definite area to talk about, or set of challenges) and for technical limitations. On one hand it's more akin to a chapter ("how long should this chapter or challenge be?"), and on the other, it's a technologically constructed limitation that hinders approaching game design with an open mind (think Dungeon Siege).

I'm not working on enough sleep. Yeah.

In the context of a very long campaign, it'd probably be best to make levels shorter; thus, the player's impression of the game as a whole would be a faster, action-driven experience. Long levels are better for more artistic, atmospheric games that focus on digging their claws into the player's psyche and creating a tangible world, rather than providing a fast fix for entertainment.

I agree that constant loading times are a continual headache... to solve this problem you could take a Metroid Prime type-approach to loading, that is, to have the next level load while the current one is being played.

Vapia makes a good point, level length isn't important as long as the player is stimulated and enjoying themselves. However, level length can be used as a tool for manipulating the player's emotional response (see above).

Correct me if I'm wrong, but don't the fluid loading systems seen in Metroid Prime, Halo 2, and Grand Theft Auto: San Andreas require a high degree of optimization and an intimate familiarity with the hardware on which the program will be run?

I can only think that it's a great challenge, and so I advise you to go about the planning phase as though you'll need some sort of "break" for loading, even if you intend to use a highly streamlined method down the road.

Well, once your player design is completed, you can predict pretty easy either at runtime (see GTA) or by use of triggers (I think Halo fits here) where the player will go, and you can use asynchronous loading of the map section ahead of you safely and without performance hits.
Fast travel over large distances (teleporters, etc) can be handled using intermissions in order not to break the system. But I feel we're going off-topic now :)