I even get significant decrease while using it on 1080's (regular not Ti).
To get the best speed you need to oc the memory clock to +500 or more. With the EWFB kernel oc of memory is not that important. with +500 on the memory clock I get 5-7% more hash than EWFB. (1060/1070 cards)
This is how it should behave. Equihash is a memory-hard POW.
This is not very accurate but it's a good approximation:
1/10 is compute bound 9/10 are memory/latency bound.