The average video game length, for a single player storyline, is probably around 10 hours. Halo 3 was about 6; No More Heroes was about 15; Assassin's Creed was almost exactly 10. There are some amazing outlyers like the Final Fantasy series, which has clocked game times of over 100 hours for its past few releases. I've heard people lament that this is actually down from five years ago, when Playstation 2 games would average 20-30 hours.
I saw There Will Be Blood last month. It had higher production values, was more engrossing, and more memorable, than most video games I will play this year. It was "only" two and a half hours long.
At the same time gamers are complaining about the increase in game prices, and developers are complaining about the increase in production costs, video games are probably the most decompressed form of entertainment. They feature screen after screen of pretty pictures with minimal interaction, every blood splatter and stock crate carefully rendered and placed. The difference is, decompression arose in comics because it was cheap and effective. In a video game, it's neither.
Now, I understand the impetus for a long game. Games are expensive, and gamers want to get their money worth. But there are other solutions! First, games don't have to be expensive. Portal is the critical and commercial darling of last year, and clocked in around the length of a feature film. The content was polished the whole way through. Almost everyone who started it, finished it. It was only $20 (or part of a multi-game pack).
Dead Rising is six hours long (with an enforced time limit) but strongly encourages replays. Your first time through the game you will certainly miss elements, because of the sheer amount of things to see. The game can deliver its key narrative over the course of six hours, and so it moves quickly, but rewards players over tens of hours with a significantly different game each time.
These two games have something else important in common, as was pointed out to me by Ben Zeigler - in theater terms, they don't draw out their second act. In a traditional three act play - which most modern films and games are based on - the first act contains the exposition, the second the confrontation, and the third the resolution and denouement. These are split about 25%, 50%, and 25% respectively in film, but more often 5%, 94%, and 1% in games; the resolution is rarely more than a final boss battle, sometimes followed by a non-interactive video. Portal and Dead Rising both adhere closely to the filmic proportions, as do other games like The Darkness which has a lengthy playable exposition.
Another option is to take the approach older games did, which has less to do with balancing the narrative proportions or replayable content than making the game really hard. Contra 4 and Castlevania: Dracula X Chronicles brought this kind of play cycle back to store shelves recently, although certain genres like bullet-curtain shooters kept it alive in modern indie games since the 1980s. The games take very little time to play through, but require playing the levels over and over to memorize enemy positions and optimize plans and reflexes. The recent release of Triggerheart Exelica on Xbox Live Arcade to complaints of "I beat the game in less than an hour [after continuing 20 times], what a ripoff!" shows that there's not universal appeal here, but the sales figures on Contra and Castlevania indicate there's still a big market for people interested in high score tables. If people are having trouble they can always turn the game down to "easy" mode, at the expensive of some points and ridicule from friends.
There is a combination of both, the path taken by roguelike games. They are hard; not brutally hard on memorization and reflexes like Contra, but strategically challenging. Combined with the fact much of the content in roguelikes is procedural - randomly generated and different each game - an individual game may take a few hours (or less), but the player will play it over and over if your design is appropriately balanced. The fact you are already generating things procedurally means roguelikes are often amenable to straightforward mathematical balancing.
There is a final option I don't need to say much about, which is to ignore single player games. Few people talk about "finishing" an MMO or a game like Winning Eleven or Smash Bros. But the single player experience is far from dead. Bioshock and Mass Effect and anticipation for Metal Gear Solid 4 suggests this isn't the whole future.
All these options succeed in getting more people to finish games - which is to say, more people hooked all the way through your game, and at the same time giving a richer experience to players who want to put more time in.
What's a bad way to handle this? Look no further than Final Fantasy - any of them, although the latter ones are bigger offenders. These are the games where 95% of the time is in the second act, and even if you ignore sidequests it takes you dozens of hours. I won't rag on FF though, since that seems to be popular in the West these days. Instead, I'll beat on Mass Effect for the same reason. Optional sidequests are not the way to make a short main narrative but a longer game. First, the sidequests are content I am interested in, but do not differ significantly from the game mechanics of the main plot. The content could easy be worked into a generic corridor in the main plot instead of a generic corridor in a sidequest. Second, they do not change the experience, they just extend it. When I play Dead Rising through a second time, five of those six hours may be entirely new to me depending on where I go in the mall. When I play Mass Effect a second time, I have the same ten hours I already did, interspersed with another five new ones. Why can't I just play the new ones?
(There is actually a solution to this, which is to write a complex multilayered narrative like those which reward repeated reading of books or repeated watching of films. Amazingly, I think Final Fantasy X did this! But then the game is so ridiculously long it cancels that out. Most of the Metal Gear series succeeded too. No More Heroes is the best recent example; the second time through the game the dialogue and camera work is identical, but knowing what you do then it's a wholly new story. But I don't expect the game industry to suddenly figure out good writing overnight, nor am I sure I want it to.)
There is still room for the hundred hour epic game, just as there is still room for George R. R. Martin books or The Cremaster Cycle. Games like Disgaea or Harvest Moon are long because the goal of the design is to have the player become intimately familiar with the game, either mechanically or narratively. But these are the exceptions, not the rule. There is not a sufficiently large payoff in putting that much time into Mass Effect's or Eternal Sonata's systems, so I don't do it, or do it only grudgingly.
When you're planning the length of your second act, ask yourself if there's really forty hours of things to do, or if you just want the player to spend forty hours doing things. If it's the latter, they most likely won't anyway.
[Since I started writing this, Leigh Alexander published a similarly-themed article about "Completion Anxiety Disorder". She comes to some very different conclusions than I did, and looks at it from a player-centric rather than a developer-centric perspective.]