lol..

-w 256 is the correct work unit for ati 5870/5970/6950/6970/6990.
tweaking the -I value.
dual cards = -I 9 (2x 5870, 5970, 2x 6950, 2x 6970, 6990)
single cards = -I 8 (5870/6950/6970)
next thing to note.. windows you'll need 1 core per a graphics card, this is due to the pthread implementation.
CPU mining is not viable with cgminer.. with a highly optimized cpu specific build.. it is still 50% slower than ufasoft's miner.
Also.. MH/s rate means nothing.. you make your "money" on shares accepted per 1 minute eg. U: 10.77/m]
cgminer will upload more shares in a given time frame than all the other miners, do a proper test.