82% CPU: Quite high. Checking the sysstat -M output it looks balanced around processing functions
183 MB/s reads + 16 MB/s writes = 199 MB/s network xfer: You are using at least 2 x GbE links. You might hit link bottleneck if traffic is not balanced.
861 IOPs: take the 199MB/s / 861 IOPs = 231 KB / op. So you are doing large IOs.
78% disk busy: Above 50% and you can already start to see some variability in latency. At 78% average the disks are quite busy; they are processing at least one xfer 78% of the time so likely some IOs are also waiting to be processed at times. If IOs are waiting at the disk layer these will eventually make IOs wait on the host side.
Disk IO efficiency: Disk read is 190MB/s, network out is 182 MB/s which is roughly balanced. Disk write is 18MB/s, network in is 16MB/s which is roughly balanced; the extra little bit is parity. You have no evidence of freespace fragmentation.
Reallocate is useful if you have freespace fragmentation (you don't) or if you have sequential read after random write (you seem to be doing large block IO so I doubt it). Even then, it is has a bigger impact on reducing disk busy than cpu busy. Checking your disk IO efficiency however I'd say it looks pretty good; network in/out matches disk in/out.
So across the main system aspects (CPU, Disk IO, Network) on each you are nearing bottleneck. The solution is to stay as-is and be satisfied with latency you have, reduce workload (or move to the other head if it is less busy) and see latency improve, or buy a more powerful controller.