That feature is actually already implemented and has been implemented from release, so you're saying you've encountered a bug with it?

When using poclbm, it invalidates a pool for 10 minutes if it receives 8 "Problem communicating with bitcoin RPC" errors in a 30s period. With Phoenix, it invalidates if it receives 3 "Failed to connect, retrying" errors in 40s. However, these checks are only made for Miner0 as they are meant to detect network errors (since everything is running on the same machine and on the same connection).

If Miner0 is working and the rest aren't it's very likely that the problem lies elsewhere, not with the pool or the network. They could have indeed crashed or have some other issues, such as bad miner accounts (be very careful with these, some pools accept shares even if your accounts are wrong, so in essence you'd be mining for nothing). If you have wrong accounts for certain pools and certain GPUs, those may not work on those pools but would be perfectly fine on others, so they would seem to stop randomly.

Also, please check actual GPU load to make sure they aren't working, that's a sure-fire way of knowing.