I am getting with the -L2 -t30x16 around 121khash/sec was previously at -L2 -t30x32 at 277khash/s.
May-be dividing the second number by two will work (autotune is certainly a better option)
I was being dense...I hadn't rebooted since the cudaminer got a bunch of "does not validate" from mining when the nfactor changed, now that I have done so the hash is looking much more healthy!
The best T autotune (T12x20 - 106kh) still marginally beats the best t one (t120x6 - 104kh) though (both with L2). My best speed is with exactly the same kernel settings as before, but with a lookup gap of 2.
I've still lost 65% of my 780 hash rate with the nfactor change, but that's much better than losing 80%.