Drew, I have the same concern. I've watched the [r] tag community grow on Stack Overflow since the early first "flash mobs" where the site was seeded with good questions and answers. That effort even got mentioned as a "good thing" by Joel and Jeff in the Stack Overflow podcast. It took a long time to get enough good questions and answers on SO for it to be a useful resource. It's also beginning to bubble up good R results in Google, which is very valuable.
Being an economist I tend to look at Q/A sites as a market. For the market to clear there has to be 1) good questions 2) good answers 3) long term providers of 1 & 2. If any becomes absent then the market fails to clear. I define "clearing" loosely as "questions get reasonable answers in a reasonable amount of time for the majority of users." Some of those users won't have to type in the question because they come to the site, find the answer, maybe vote an answer and/or question up and then they go away with what they came for. That market just cleared. What has helped SO become a clearing market for [r] questions is the concentration of many good questions where lots of folks find answers before they type questions and lots of fast answers when users do type in a question.
So coming back to Drew's comment: If the set of R users (which is very small compared to Python, Java Script, PHP, etc.) gets split across multiple Stack Exchange sites I'm concerned that the R questions on all of those sites will fail to clear.
If I was "King of the Internet" (registered trademark of Al Gore) I'd make all questions that are mostly syntactical go to Stack Overflow and let the ones that are more related about an algo go to a site devoted to that algo or concept. But even as King that would be a real bitch to do in practice. In SO R questions I see lots of questions that the asker thinks are about his field or his algorithm but, in reality, he's struggling with an R syntax or data structure problem. So expecting the user to understand what level of abstraction his question resides is not reasonable. So as King I would just move the question... assuming I could determine the appropriate level of abstraction, which I probably could not.
I'm concerned that expecting this to 'self organize' may not be reasonable and we may neuter Stack Overflow as a source of good R questions and answers.