I think I've just discovered something by chance that has completely floored me.
I have a bunch of policies with a primary dedupe disk copy and a pair of secondary copies going to tape.
I've always started an Aux Copy with the defaults i.e. all copies, maximum streams, which meant that for each policy running an Aux Copy fired up a single auxcopy.exe process on the MA with two streams, one to each tape.
Performance was all over the shop, very unpredictable and quite low - nowhere near the throughput that the disk or both drives can support.
It seemed obvious something wasn't right as doing a data verification job on the source job using the same streams was giving phenomenal throughput.
So, I just tried something which was to start two aux copy jobs for the policies, one for each secondary copy.
The result is two auxopy.exe process running, and my tape drives look to be running at full speed.
To say I'm staggered at the difference in performance is an understatement - it's not something I've read about before and I'm assuming it's related to how Commvault managed resources per process.
Hopefully someone will find this little tip useful.