Ours set to 5.
I looked at concurrency more in terms of balancing disk activity with network utilization. In our case on average our 3 dedicated gigabit network connections are 90% saturated when concurrent jobs are set to 5. The jobs are split up in fast and slow groups across the network interfaces. Jobs are also staged where slow systems all run at the same time on one interface and fast systems occur two at a time each on their own dedicated interfaces. This allows two fast systems to saturate two of the interfaces while 3 slower systems run concurrently to saturate the 3'rd.
If our network had more bandwidth, increasing concurrency would help, our disk array has a lot more speed in it.
All the jobs are set to run the verify stage when backups are not occuring, I notice even a single verify job eats up 100% of the disk array.
