I just got 2 s7's and setup one so far and after 20 minutes it has over 5k errors....nearly double your errors in a fraction of the time. Im kinda scared to see what issues the second one has. The first one also hashes at 3.2th/s and one chain runs way hotter then the other 2 chains do. I emailed bitmain though so hopefully they know what to do about this.
Sometimes when these rigs start up they can "get up on the wrong side of the bed". Sometimes restarting the mining will fix these problems. Other times you will have to reboot the node. Other times you may have to power down the node and wait a while and try again, in which case I have found that it will usually be necessary to reboot them once and then restart CGminer as well. I suspect that there is hidden software/firmware that is turning off cores if there are too many errors, and this may result in poor decisions if early operation is unlucky and hits a number of hardware errors. There may also be problems with how the software computes average hash rates.
For best results, you will have to experiment with clock rate. The effective hash rate is not a simple function of clock rate times (1 - hw error rate). Also, you want to set manual fan speed and try to keep temperatures in the low 60's for best results. Or you can just accept that some machines are better than others and down clock them a bit and not sweat the small stuff. I am getting pretty much what I paid for in total, but two of my three machines are slightly over clocked and the third is slightly underclocked. My main daily concern now is keeping the temperatures in the desired range and enjoying my free electric heat.
