Recently, I gave some reasons for SIAI to begin publishing in mainstream journals, and outlined how it could be done.

I've recently been made aware of some pretty good reasons for SIAI to not publish in mainstream journals, so here they are:

1. Articles published to websites (e.g. Yudkowsky's work, Bostrom's pre-prints) seem to have gotten more attention, and had more positive impact, than their in-journal counterparts.
2. Articles in mainstream journals take a relatively large amount of time, money, and expertise to produce.
3. Articles in mainstream journals must jump through lots of hoops - journals' aversion to novelty, reviewer bias, etc.
4. It is easier to simply collaborate with (and greatly influence) established mainstream academics who have already jumped through mainstream academia's many hoops (as Carl Shulman has been doing, for example).