I think you need to add a fourth option: People with a blatant conflict of interest--most often a political affiliation. Even Wikipedia (which thrives on niceness and NPOV-seeking, if not truth-seeking) assumes bad faith when people try to edit their own articles, with its WP:COI policy. The legal system assumes bad faith about prosecutors and lawyers, which is why its standard of evidence is so extreme. And while the scientific method does not assume bad faith about scientists, it still protects against their naive errors of rationality, which is just as important.
Of course, we already know that prediction markets excel at integrating people with bad faith in a rational, truth-seeking institution; finding a way to do the same thing in a deliberative forum comparable to Less Wrong would be an extremely useful development. My hunch is that it would be useful to steal a page from the playbook of politics and support clearly-defined factions with different points of view and perhaps different policy proposals or decisions or what have you. But all of this is largely speculation.
tl;dr: I think Alicorn's post is definitely cogent when it comes to LessWrong as we know it. But there's a huge design space to be explored for more resilient institutions.
Hrm. Well, if politics itself is any example to judge by, that may make for a resilient institution -- but the mess of allegiances and biases created by splitting people into well-defined factions probably entails that the institution would be much worse off in terms of truth-finding, because devoting too much of its energies to internecine squabbling.
I suppose you need to strike a balance between unproductive antagonism, and ending up as a group of like-minded folks just patting each other on the back. Thankfully, LW seems to have a strong dose of "Let's get to the bottom of this"-type norms, and the appropriately rigorous/persnickety personalities, to stop it from getting too back-patty.
tl;dr: Sometimes, people don't try as hard as they could to be nice. If being nice is not a terminal value for you, here are some other things to think about which might induce you to be nice anyway.
There is a prevailing ethos in communities similar to ours - atheistic, intellectual groupings, who congregate around a topic rather than simply to congregate - and this ethos says that it is not necessary to be nice. I'm drawing on a commonsense notion of "niceness" here, which I hope won't confuse anyone (another feature of communities like this is that it's very easy to find people who claim to be confused by monosyllables). I do not merely mean "polite", which can be superficially like niceness when the person to whom the politeness is directed is in earshot but tends to be far more superficial. I claim that this ethos is mistaken and harmful. In so claiming, I do not also claim that I am always perfectly nice; I claim merely that I and others have good reasons to try to be.
The dispensing with niceness probably springs in large part from an extreme rejection of the ad hominem fallacy and of emotionally-based reasoning. Of course someone may be entirely miserable company and still have brilliant, cogent ideas; to reject communication with someone who just happens to be miserable company, in spite of their brilliant, cogent ideas, is to miss out on the (valuable) latter because of a silly emotional reaction to the (irrelevant) former. Since the point of the community is ideas; and the person's ideas are good; and how much fun they are to be around is irrelevant - well, bringing up that they are just terribly mean seems trivial at best, and perhaps an invocation of the aforementioned fallacy. We are here to talk about ideas! (Interestingly, this same courtesy is rarely extended to appalling spelling.)
The ad hominem fallacy is a fallacy, so this is a useful norm up to a point, but not up to the point where people who are perfectly capable of being nice, or learning to be nice, neglect to do so because it's apparently been rendered locally worthless. I submit that there are still good, pragmatic reasons to be nice, as follows. (These are claims about how to behave around real human-type persons. Many of them would likely be obsolete if we were all perfect Bayesians.)