Using the .error property as indicator is the only way right now. The link return doesn't (and shouldn't) know anything about the nodes connected, that would make it much more complex. I think it should only react on the message.
I think of .error as a convention for errors, just like payload for data.
I actually use it that way sometimes to generate an error that is not from a catch node, so I can use the same error handling flow path.
In some of my own nodes (request-response paradigm) I use the presence of .error to send an error reply instead of a regular data reply.
So at least for me, this convention has become a viable best practice.
Of course, using the upcoming group catch feature would reduce the effort of selection the individual nodes significantly. 
