The danger here is that the easy-to-upvote throwaway comments become a self-fulfilling prophecy. Like it or not, karma is social validation. A long, thought-out reply that gets no upvotes is somewhat depressing; it makes you think that, even at your most thoughtful, your thoughts are unappreciated. Contrast that with a cheap and easy line that gets 40 points, and the social feedback is clear: your considered opinion isn't valued, but your cheap lines are.
It makes me think that the karma system itself might be fatally flawed for its lack of scalability. What works for a small, mostly homogenous community breaks when that community becomes larger and more diverse. We can deny it all we like, but I'm sure most people feel pretty good when they log on and see their karma significantly higher than it was the last time they checked. It's an ego boost, and few people are immune to enjoying that. When playing to the lowest common denominator gives you that fix of social validation, and a more thoughtful comment doesn't, it's pretty strong positive reinforcement for the less-desirable behaviour.
I'm not sure how to combat this in a scalable manner. It's difficult to think of examples of community-moderated forums maintaining quality in the face of rapid expansion. My previous suggestions of weighting the votes of those who consistently upvote quality comments higher than those of people who upvote fluff has met with little traction, and I can see some pretty big potential weaknesses there, myself.
In reviewing this comment, however, I have had a thought of something that seems ridiculously trivial, but might just work: pg has stated before his belief that comment length is a relatively good indicator of quality. I notice, however, that the comment box is relatively tiny. It makes a long, thought-out post appear, prima facie, no more substantive than a 3-liner. I would be interested to see what the result of making the comment box significantly larger would be. Would people, upon seeing that much empty space surrounding a throwaway comment, be likely to reconsider posting it? Are there easy cues like this that can be used to hack the behaviour of commenters? I'd be interested in seeing the results of running an A/B test on a seemingly simple change like this: Make half the users' comment boxes two or three times as long. Leave it that way for a few months, then take a random sample of the resulting comments and see if there's a difference in quality. I wouldn't be surprised if there was.