There are two competing forces that we have to strike a balance between:

1. It's good to be pretty specific. That way the people who visit the site have a good reason to visit it. They don't have to dig through a long list of questions to find something they're interested in, so the site is actually usable.
2. It's good to be pretty broad. When somebody comes to MO, they should get to ask and answer detailed technical questions in their area, but they should also have their curiosity piqued by questions in tangentially related areas. It's fun to see what people in other fields think about, and it's awesome when it turns out that somebody in a different field has exactly the right tools to solve your problem. Breadth of topic makes MO more fun and more valuable.

I think (2) is strong enough that we should shoot to be as inclusive as possible so long as we don't end up with a fragmented community. The situation we want to avoid is having two or more essentially non-interacting sub-communities on a single site. They would just annoy each other since they'd have to go to the trouble of filtering out what they want to see. The communities would be much better served by separate sites. Let me add a couple of clarifying points in anticipation of misinterpretation:

• "But the MO community is already fragmented; I'm mostly only interested in combinatorics and I'm never going to read algebraic geometry questions." It's true that individuals are going to have some things that they're really interested in and some things that they're not interested in at all. That's why we have interested and ignored tags. The point is that there are enough people interested in both combinatorics and algebraic geometry that the benefit they get out of having both in one place outweighs your irritation at all the algebraic geometry questions. True fragmentation of the community would be a situation where the "common interest graph" can be cut into nearly disconnected components. Not having every person interested in every topic doesn't count as fragmentation.
• "But you don't allow vague questions, even though people are interested in them." Part of the reason for that is that the Q&A format and the SE platform is bad for vague questions; see my other posts on the issue for more detail. But I would also argue that the me who is interested in posing and solving focused problems should be treated as a different person from the me who is interested in having a discussion about some vague or philosophical stuff. Of all the things I'm interested in, I'm not interested in all of them at the same time, and me in one mood may be thoroughly annoyed with me in another mood. From this point of view, I think mathematicians in the mood to solve focused problems and mathematicians in the mood to chat philosophically constitute fragmentation of the community. But that's okay; it just means that you should have separate sites for the two moods. Just because you really like to be on MO sometimes doesn't mean that you have to satisfy all of your interests on MO.

In the case of computer science, I think CS people should experiment. For questions related to programming (e.g. any question that involves a specific language), ask and answer them on SO. For theoretical questions that you could classify as research/graduate level, ask and answer them on MO. If it turns out that people at SO and MO are happy to have those questions and this covers most of your field, great! It means that you only have two websites to keep an eye on, you get the advantage of a large number of eyeballs, and you make SO and MO better. If it turns out that people at SO or MO don't like it, or that you have questions that don't really belong on either one, then it may be worth the hassle of creating and running CSOverflow.

Of course, there's a continuum that runs from pure mathematics to hardcore computer science, with theoretical/mathematical computer science lying in between. There's no obvious place to draw the line. If we do manage to attract more applied mathematicians, we'll have to fine-tune our instincts as to which questions to allow or encourage and which to declare to be outside the site's remit. But that's OK; there have already been several issues on which the community's had to feel its way.

Afterwards, there will be a natural overlap of questions between the math site and CS sites, which is fine, and which is exactly as things stand among published papers in the two subjects.