Thanks for the clarification. It's a pity that you have to drive into the context sensitive help to understand how it really works.
In our case, we have servers that are using FC to talk to the Storeonce, and a Tape writing server to write tapes. So for object copies we want the tape writing server to read from the Storeonce over FC, and write to tapes over FC. What we started seeing in our copy jobs, was that the copy job was using the original server to read from the StoreOnce over FC, then sending it to the tape writing server over ethernet, which was then writing to the tapes over FC. And the ethernet part was a significant bottleneck. It doesn't help that the source servers are HPUX Itanium, and the tape writing server is Redhat x386 ...
Changing this behaviour using the Source tab in the object copy spec seems to have worked.