Why don't you try out new pools without disconnecting from the current one ? I've tried that and it works well for both windows and linux miners (using rpcminer-cuda and phoenix for two different rigs). Would at least solve the problem of idle machines if something goes wrong while giving you the same data about reliability of the new pool.
I did this when I was testing around, but when I tried to make the full switch from BTCmine to BTCguild, that's when the problems started.
Why don't you sSplit the cards between all the US btcguild servers. You need to use 1 worker Per GPU core.
That's a great idea, I think that might actually help. I will definitely try it out!
I think it is important that there is some way for shareholders to verify the hashing power of the cluster, so I am planning on having the computers back on a pool that allows for this by the end of the weekend.