Timothy Farrar shares some interesting thoughts on parallel rendering.
Timothy points out an article parallel SPU rendering from GDC08 by Mike Acton of Insomniac Games.
This article has some insight turning conventional wisdom on its head. In particular, I am struck by the section on rendering fractured glass where the first approach was to take a high level view of fracturing the glass, then shove it down the pipeline. The second approach was to design a pipeline that could handle dynamic geometry, then fit the approach to the glass to what was achievable.
Old school, that's how we approached things. I feel this philosophy is a bit lost these days; top down is so ingrained in the way we are trained to think as modern software engineers - but Insomniac's results speak for themselves.