Neither. I've never done any large-scale coding in Scheme (I'm a Common Lisper) but it's more than just a hunch. The theory behind name capture is easy to understand. The only counter that I've heard is, "Well, it just doesn't seem to be a problem in practice." I believe that this is only because the Arc code and user base have not yet reached the scale where the problems manifest themselves.
But even if Paul is right and name capture is rarely if ever a problem in practice, punting on name capture seems to me fundamentally at odds with the stated goal of producing the 100-year language. At best it's an unnecessary burden on the programmer, and at worst it's a time bomb that could go off any time. Either way it doesn't make me optimistic that Arc is going to win in the long run.