Basically, if you look at the architecture of any modern GPU, R5xx/6xx or G80, it comprises pretty modular units connected by a big interconnect. Imagine if the interconnect was more distributed like say an Opteron and [HyperTransport], you could have four small chips instead of one big one.Such a design would be cheaper to manufacture, since a smaller die area per chip could reduce the likelihood of defects. In fact, Intel already uses a similar dual-chip technique in its first quad-core processors. This more modular design approach could also conceivably help ATI release slower derivatives of its high-end products sooner, should it use the same chips across multiple product lines.
This would have massive advantages on design time, you need to make a chip of quarter the size or less, and just place many of them on the PCB. If you want a low-end board, use one, mid-range use four, pimped out edition, 16. You get the idea, Lego.