My opinion is that there is no point in having STOC/FOCS special issues at any journal devoted to theoretical computer science at large. My point is that the scope of STOC/FOCS is almost identical to the scope of such a journal, and so there is no added value (for science) in having such special issues. (Of course, it may be the publishers interest to have such a special issue, but I do not see why the scientific community should contribute to an effort that only benefits the publisher.) Similar considerations apply with respect to having a special issue of, say, the Computational Complexity Conference in the Journal of Computational Complexity.
Special issues just generate a host of administration problems, and the only reason to have them is if they serve some good cause. For example, a special issue on some area is a good way to focus attention to that area. Likewise, a special issue to commemorate some special event. None of these applies to a STOC/FOCS special issue (in a journal like JCSS or SICOMP).
I don't see the point of having a special issue that contains a selection of 7-15 papers out of 60-70 papers presented in a given FOCS/STOC. Furthermore, the presumption and/or pretense that this selection reflects the "best papers in the said conference" is harmful as well as highly problematic:
(Still, I do not consider this issue to be that important as to justify my resignation from the editorial board of journals that decide to have special issues of the aforementioned nature...)
Back to Oded's page of essays and opinions or to Oded's homepage.