Martin Pool links to a usenix paper from last year on Crash only software. It is an interesting paper, the first I have seen applying this principle to J2EE. Still I was rather disappointed that the only prior work on supervisorary fault recovery cited seem to be their own. For instance the erlang community has made this a fundamental part of their culture, and the approach is pervasive from the language primitves thru standard libraries to system architecture recommendations and community norms.
Anyone who might be interested in more detail on how this philosophy works in practice is encouraged to check out Joe Armstrong's PhD thesis Making reliable systems in the presence of software errors.