That's weird, considering I see the exact opposite opposite results when I lower -f. If i set it at 1 the computer becomes hard to use due to graphical lag, but I get an extra ~10% Mh/s.
What if you we're to run the cards independently with -f set to 1?
355-357 MH/s each, a very, very slight improvement. A bottleneck is created somewhere when -f parameter is set very low while CFX is enabled. I'm not knowledgeable enough to know what's going on, just experimental data to suggest. Others experiencing significant slowdowns on CFX should try -f 50 or higher, or just remove the argument altogether and see if it improves.
With -f 1 with CFX enabled, one GPU is @ 96-99% load while the other is low and extremely erratic, sometimes only 2 MH/s, then next poll it's at 250 MH/s. With -f 50, or without the argument altogether, load on both cards is 99%.