I have been experimenting with XFS on two systems. One has 8 virtual cores and the other has 12. I move a lot of large (2-20GB) files back and forth across a gigabit LAN between these two machines. Whichever machine is being written to during these directory syncs becomes very unresponsive. HTOP shows that nothing signifcant is going on in terms of CPU or memory usage so I'm really not sure what is going on. One of the machines has XFS and ext4 partitions and it doesn't seem to slow down, but that particular machine has ext4 on RAID0 and xfs on RAID1 which isn't the best of comparisons, but both machines do this. I am using stock 14 with huge kernels on both machines. Is this something that just happens if xfs is used or is there something I could do to stop the sluggishness?

I saw this when doing things like copying the linux src directory from one place to another on an ext4 on lvm on luks setup. I found running the workload under "ionice -c 3" seemed to resolve the situation. Interestingly when I converted to a ext4 on luks on lvm setup the interactivity hang-ups seemed to go away whether I used "ionice" or not: which kind of suggested to me that the device-mapper i/o scheduling stuff isn't quite working correctly. But this is all anecdotal, I have no evidence at all to back this up.

You could try ionice, failing that, you could try swapping from CFQ to the deadline scheduler and see if that helps.