This reminds me a bit of the K/kdb approach of just starting another process for almost certainly the same reasons.