A very interesting thread, and I'll sum up my responses to the various points mentioned in this thread:
Dungeons in general: Some posters have mentioned how dungeons run the risk of becoming stale and stagnant if they're just underground constructions of rooms and tunnels, but I think the typical dungeon crawl can be something much different.
Just look at some of the original Dragonlance modules. The various dungeons were actively incorporated into the storyline and history of the campaign setting, and the players went in for reasons other than staggering out under loads of treasure-they had to enter Skullcap to get information on how to find the Dwarfgate, needed to enter the Nightmare of Silvanesti to retrieve a Dragon Orb, enter Pax Tharkas to rescue the prisoners, enter the Glitterpalace or the High Clerist Tower to get valuable divine aid in their struggles-and in every case these dungeons were given backgrounds and reasons for existing, and woven into the storyline. Sometimes the PCs could get a wagonload of treasure, but what could they spend it on? There were other reasons for getting it.
The earliest modules were done to help out DMs who hated dealing with the drudge work. Speaking as someone who has absolutely no artistic talent, it would have been a relief to have the dungeons designed for me if I were a DM in the 1970s (I wasn't even born until 1982, so I'm only 24, yet I love the feel of 1E-go figure). Gygax pretty much assumed that DMs would rewrite and retool the available modules to their own campaign needs, saying so both at the beginning of the GDQ modules and even in the 1E DMG itself.
The original giant holds could serve any number of other purposes. Are your players in a long-running feud with the Zhentarim, who are forging alliances with different types of giants? Well, Nosnra, Snurre and Grugnur could easily be recast as allies of the Zhentarim, their fortresses fitting comfortably into your players' plans for battling these enemies of the Zhentarim.
Are your players, in seeking out the aid of a northern barbarian chief, made to rescue his captive son, wife or daughter from frost giants in exchange for help? The Glacial Rift can easily be re-used as a one-shot adventure where the players have to accomplish this task, an interlude in the task that allows the players to get back to the regular campaign at hand if they succeed-they could refuse, and in that case not have the aid of the barbarian, but they could find allies elsewhere...
Do you just want a night of hack and slash, with no greater evil involved than a marauding hill giant chief? Chuck out the references to the drow, and the magical chain, make Nosnra himself the major villain, and cut loose! There are no grander plots, no alliances, just a mob of giants causing havoc-and a party of bold adventurers that have to stop them dead.
2E modules like John Terra's "Four From Cormyr" had both background and dungeon action, but Terra also included a wide variety of loose ends that DMs could exploit to bring up further adventures, and his adventures also contributed to Faerunian history, with the exploits of the Starburst Swords and the ill-fated Kingdom of Orva. They were four separate, but interlocked, plotlines that DMs could incorporate when the time seemed right, but there was no need to do them each one after another. Some of them were more plot-driven than others; in some cases, the plot was not so integral that the DM couldn't change the location or characters involved with a minimum of fuss.
Some of the new modules I've seen, most notably the Kingdoms of Kalamar ones, are I think an interesting mix of dungeon crawling combined with role-playing, with the dungeon incorporated as part of a greater plotline, and one that leaves player discretion in some cases-again, taking the cue from Dragonlance to have going into the dungeon for greater reasons than simply getting treasure.
I suppose the nutshell is that over the years, steps have been taken to weave dungeon crawling in with the plot, and give players more reason to do it than simply gathering up masses of gold and silver. Dragonlance was one of the first to do this, and they've taken it a step further.
At the risk of being accused of self-serving, many of my own "Newly Discovered Dungeons of the Flanaess" articles from Canonfire have tried to come up with more original reasons for the PCs to enter these dungeons, and to let DMs incorporate them into their own campaigns:
http://www.canonfire.com/cfhtml/modules.php?name=News&file=article&sid=347
http://www.canonfire.com/cfhtml/modules.php?name=News&file=article&sid=350
http://www.canonfire.com/cfhtml/modules.php?name=News&file=article&sid=735
Some of them respond, for example, to specific metagame needs: Sometimes PCs can die in ways that make resurrection impossible-being destroyed by magnesium spirits, devoured by changelings (or some other obscure 1E monster), turned into horrible undead wraiths and shadows, devoured by a demilich to fuel its life energy, etc. Well, as long as the player is remembered, the players can travel to the Valley of Memory to find a way to restore him. It's also meant to be a dungeon you have to _role-play_ through, instead of just hacking and slashing.
Science fiction plotlines sometimes have characters shrinking down to go inside a person's body to save them. But what about going into the insides of a demon to destroy a demi-god's divnity? If Iuz is your campaign's major villain, the Balor's Stomach is the dungeon for you.
In the Sunken City, characters who persevere may be rewarded with the revival of the old Flan kingdoms, which can lead to all sorts of political or personal complications-role playing, anyone?
I tried to come up with more rationales for players to actually go into these dungeons of their own free will-if you as DM make the knowledge available to them, the players can follow up on it. If they're not interested, then that 's that; but if they are interested, then you can write it up for them, and there's no railroading involved.
I also noticed a fair bit of criticism of the Dragonlance modules-while I can understand it, I think that DL is a special case. If they're being played properly, your players are taking up the roles of characters who already have their own personalities and backgrounds, and who have a vested interest in finding the True Gods. Now, when two plains barbarians show up with exactly this evidence, they have motive for following up on it. One thing leads to another-if the players are expected to go to Silvanesti to stop the Nightmare, they have reasons for doing so because at that point they think the Dragon Orbs can destroy the dragons, and they have a chance to gain one, just as the other party has reason to go to Icewall. The players have a reason for going to Sanction to rescue the dragon eggs-getting the good dragons involved in the war.
I think that the unique situation of the first Dragonlance modules comes from the fact that there are pre-written characters who already have their own stories. I doubt many DMs or players would have had as much of a problem if these characters were ones they had created themselves, and had developed for themselves the interest in the True Gods, in the fate of the elven nations, or the interest in the fate of the dwarves or the Knights of Solamnia. Rather, I bet the players would have been eager to follow up on leads that fed into their characters' backgrounds and personalities. However, since the whole thing was a commercial series of modules, it can't help but feel like a railroad in some cases.
Even then, Dragonlance is not a completely rigid track-the recent 25th Anniversary re-issuing of the Dragonlance modules by Steve Miller and Stan! offer a useful story frame and guide for DMs. If you use that to create potential variables and add in interludes where the players can muck about in their own way, it feels like less of a railroad. If the players have to choose between helping the Knights and go to Icewall, or help Ahlana and go to Silvanesti, and follow up on one or the other of the leads later, or openly decide to divide their forces, there isn't as much railroading.
As a sidenote, I vastly prefer the "module" version of Dragonlance, mostly because it doesn't cut out so many interesting scenes (King of the Deep, journey to Sanction, the Glitterpalace, the possibility of using Huma's Lance to seal the gate at Neraka, which is how Sturm OUGHT to have died, the journey to Skullcap, etc.) and doesn't turn certain characters into ballast that play a minimal role in the story (coughGoldmooncoughRiverwindcoughTikacoughFlint) suffer from deus ex machina (coughKitiara showing up at the last minute to bail the Heroes out when they get busted by securitycough), or have completely unnecessary, arbitrary and ignominous deaths (coughFlint's pointless death)...
...but that's another rant.