I find the problem of community dilution very fascinating. It does not only apply to news sites, but also to political parties, religions, groups in general. Probably there are tons of academic papers about it available somewhere - please post links, if you know any interesting ones. Unfortunately I don't know any :-(
However, I just had a thought for emulating the problem:
Create a random set of ideas, and give the simulated users random subsets of the ideas. Assume the users are undisciplined. For example the ideas might be "startups, hacking, stupid cartoons", and you don't really want to have "stupid cartoons" in your hackers community. But because users are undisciplined, they will upvote "stupid cartoons" anyway (if it is in their interest), even though they know they shouldn't.
Now you could go ahead and submit random stories (ie with random idea subsets) and see what happens to the news page. For example start with a high proportion of users who are only interested in hacking. Make up rules for their behaviour (leave site if signal-to-noise below x) and so on. Experiment with the results of different reward systems etc.
Hours of fun - I wish I had the time for it... Perhaps with a classifier, it would even be possible to analyze the real Hacker News.
