Ok, cools. But say MS wanted a certain amount of performance. Let's say to get there it would require whatever number of whatever clusters. Altogether you need 6 billion transistors to get all the clusters you need. Do you make a single 6 billion transistor chip, or do you make two 3 billion transistor chips? Now, with traditional SLI and Crossfire you do not get a linear increase in performance (2 GPUs does not constitute a 100% performance increase), but perhaps with what they describe in the op, they are implementing something that gets closer to a 2 x increase?
The logical thing to do would be to create a very nice, very efficient brand spanking new GPU. But between what Alberto said and this rumor, I'm starting to think its possible they are going for two GPU's to get a larger than expected generational leap. Then again perhaps IGN was more on point and its a mild increase, or perhaps something closer to the Orbis rumors. Something more traditional.