Actually, one thing someone was saying about the possibility of dual-channel SDRAM systems. Surprise! they already exist. ALI created a dual-channel SDRAM interface (128bit wide) for the socket-7 market. They put it in place mainly because it was integrated with the video also, so that the performance benefits weren't all that hot because the main memory bandwidth was eaten into by the video bandwidth.
But this proves it is certainly doable. It was reviewed @ Tom's a while ago. It would be really interesting if ALI was to apply this to their coming DDR chipsets. 4200mbps, low-latency memory interface, would be terrific.
edit: Total speculation, but I wouldn't be surprised if Intel did this for the P4. dual-channel sdram would provide the same bandwidth that their current rambus solution provides, but at a lower latency. If Intel does this, then RDRAM is truly dead on the PC for the next few years, at least.
Regarding rambus's fundamental super/inferiority, one thing to remember : Alpha went with rambus for their new chip because of its extraordinarily high bandwidth-per-pin. With an 8 channel rdram interface, they get massive bandwidth, and by integrating the ram controller onto the chip itself, latency goes way below SDRAM+discrete controller. If SDRAM was integrated it would have a similar benefit, but single or dual-channel SDRAM would add too many pins to already complex processor pin-outs.
So serial ram might have something going for it in the future. But if so, I'd prefer it to be an open standard, rather than Rambus's proprietary design.