We've got a single non-responsive load balancer IP in one of our primary ELBs (others have been fine for several hours now), so while everything else for us is up & running, still have transient errors for folks that get shunted to through that one system.

The interesting thing about the ELB in a situation like this is that I believe it may, in many instances, be better to hobble along and deal with an elevated error rate if at least some of your ELB hosts are working than to re-create the entire ELB somewhere else, especially if you're a high-traffic site where you may hit scaling issues going from 0 to 60 in milliseconds (OMMV, but we've been spooked enough in the past not to try anything hasty until things get back to normal).