Have you ever thought about to integrate the Netflix/Hystrix framework with the CA API API gateway?
One idea would be to integrate it with the routing assertion to detect bad behaviors of backend servers. Instead of trying to route hundreds of request to a backend which is for example overloaded and not able to answer requests (or only very slowly), you could react proactively in the gateway, e.g. not route the requests to overloaded backends.
The gateway could be protected, the wast of gateway resources could be avoided, e.g. allocating and block connections and memory.
On top, load could be taken off from already heavily stressed backends, they had a chance to recover even earlier.