Kaveh asked what problem the two level split into research and non-research sites would cause. This has now gotten a couple of answers. I asked what problem is solved which I think is more important, and Kaveh thought this would be a different discussion, so here it is.
The only argument I've heard is that you need separate research sites to not waste the time of experts, but that is a strange statement, answering questions is almost by definition time wasting, you do it because you enjoy it, not because you have to or make money from it.
And in any case experts generally do not have the time to sit and pounce on questions when they arrive (although there are some experts like that on Stackoverflow too) instead you will check very narrow special interest tags, or you will look for questions with bounties or questions other people haven't been able to answer. Simple questions will be answered by others.
So the claim that an all-inclusive site would waste the time of experts is clearly incorrect. Each expert chooses how much of his time and one what he wants to waste. The existence of simple questions does not force anyone to answer them.
So: What problems does the two-level split solve?
Update:
We have been discussing this for a while now in several threads, and I think we are starting to run out of steam. My conclusion is this:
The problem a two-level split solves is that some researchers doesn't want to hang on a generic site. The arguments for why are to a large part clearly just rationalizations of their feeling on the subject, but some arguments make sense, like that the research-level questions tend to be more discussions than answers, and hence are a different type of questions. But rational or not, the opinions of the researchers should be acknowledged.
In my opinion it would have been better to start with a generic site, and split out a research-level site if there seems to be a demand for this (ie research level questions), but not enough supply (ie answers), although this is of course too late with cstheory.SE.