This is a rule of thumb when using identical cards, but the eyefinity6 will have 2GB, and the 5870 will have 1GB. Your rule would imply that the 5870 will be able to reach across the PCIE&crossfire connection to access 512MB on the eyefinity6's pcb. My question is whether that is possible, and whether it will be a functional limitation.
ha ok didn't understand what you meant the first time
well still the answer is "nearly" in my bit of text
each and every GPU need to store a local copy of the first GPU's memory
if one GPU doesn't have enough local memory then each & every GPU is limited to the smallest amount of local memory available to any of the GPU's at hand.
Bandwidth & latency constraints are the reasons why it is like that. Storing data needed for GPU calculations on another card would cripple performance.
so with a 1GB card + 2GB card, in CF you'll get 1GB of usable memory/framebuffer