Spidermonkey lets you set a recursion limit so you don't blow out the stack and coredump. Their shell does this but we're not setting it. It's particularly important because they also don't do cycle detection in their JSON routines at this point, so you give it an object with a cycle and, bam, segv.
They use 500KB, which gives me about 1500 stack frames in the JSON case.
Works for me unless anyone else has an opinion.