I am surprised that the contact is so poor even with the stock heatsink that should hold the heatsink down pretty good. Do you know if the heatsink is bent or it is the FPGA?
I've tried another heatsink (Alpenföhn Ötzi). Result: same contact pattern, so the FPGA-4 is bent. Checked FPGA-3, bent too. I've put some extra grease in the middle. Frequency went up to 200, but with errors.
Stats without the "extra" amount of grease in the middle:
005-0: ztex_ufm1_15y1-04A3476278-1: f=200.00MHz, errorRate=0.02%, maxErrorRate=2.60%, hashRate=200.0MH/s, submitted 15 new nonces, luckFactor=0.99
005-0: ztex_ufm1_15y1-04A3476278-2: f=196.00MHz, errorRate=0.09%, maxErrorRate=1.91%, hashRate=195.8MH/s, submitted 13 new nonces, luckFactor=0.99
005-0: ztex_ufm1_15y1-04A3476278-3: f=192.00MHz, errorRate=0.00%, maxErrorRate=0.52%, hashRate=192.0MH/s, submitted 15 new nonces, luckFactor=1.01
005-0: ztex_ufm1_15y1-04A3476278-4: f=196.00MHz, errorRate=0.00%, maxErrorRate=0.98%, hashRate=196.0MH/s, submitted 15 new nonces, luckFactor=0.99
...and after:
005-0: ztex_ufm1_15y1-04A3476278-1: f=200.00MHz, errorRate=0.24%, maxErrorRate=1.22%, hashRate=199.5MH/s, submitted 14 new nonces, luckFactor=0.92
005-0: ztex_ufm1_15y1-04A3476278-2: f=196.00MHz, errorRate=0.51%, maxErrorRate=0.88%, hashRate=195.0MH/s, submitted 12 new nonces, luckFactor=1.09
005-0: ztex_ufm1_15y1-04A3476278-3: f=200.00MHz, errorRate=0.00%, maxErrorRate=0.50%, hashRate=200.0MH/s, submitted 12 new nonces, luckFactor=1.00
005-0: ztex_ufm1_15y1-04A3476278-4: f=200.00MHz, errorRate=0.87%, maxErrorRate=1.78%, hashRate=198.3MH/s, submitted 16 new nonces, luckFactor=1.02
Here the photo of the other heatsink: