We always use single node for backup od database (name that we should use no matter where instance is running) and for archive logs we use all nodes separatelly.

What is the reasoning behind this? Also, is connecting to an od instance versus using the service name going to be an issue when restoring to another node?

In old environment they do not use load balancing... so specifying service name doesn't give us much. We simply allocate channels per nodes. In new environment we are going for geo-RAC and there probably we will use different setup there (currently building up). That kind of stuff is really DBA decision and most of the time they build their own RMAN scripts which we just call for backup. Restore is also something they handle most of the time. I once read their paper on how they do it and at that time it made sense... but that was few years ago...