Still doing it.
I gave them a break from WCG. No dramas on PrimeGrid or Ibercervis.
On spooling them up again just now each one generates a unique ID, but when they connect to WCG it considers them to all be the same machine with the ID of the most recent one to communicate. The ID changes each time another communicates. It's also sending all of them the exact same work units, saying "resent lost task" for work that's still running on another instance.
They were working fine until recently and still work fine on other projects, so I highly doubt it's the instances themselves or their configuration that are at fault.

Have you tried setting <allow_multiple_clients>1</allow_multiple_clients> in the cc_config.xml. This is what I use to run multiple clients on the same host when I hit limits in the client.

Exactly, it facilitates for clients to load/run concurrently on one OS, assign separate RPC ports, 31416, 31417, 31418 etc and separate data directories. If you don't set that flag, the client checks if one is already running and if so wont load.