See the comment in the upcoming patch. Attempting to "fix" this is not a good use of resources at the moment, and may lead to other bugs. The behavior triggered by bug 671196 is good: the parent process gets into a state it can't handle and kills the child process. The problem is, the reftest harness doesn't know how to restart crashed content processes. We need that, but bug 671196 is more important right now IMHO.
Created attachment 646491 [details] [diff] [review] Disable 735943.html for <browser remote>, for now See comments in patch. We should ensure that the mozbrowser window.open setup can handle this edge case. This also raises the question of how to integrate <iframe mozbrowser> into the reftest harness, which you may want to participate in. (Swapping it out for <browser remote> should be trivial.)
Comment on attachment 646491 [details] [diff] [review] Disable 735943.html for <browser remote>, for now sgtm