I wonder where you got the idea that RDRAM is superior to DDR-SDRAM. Is it because an i840 chipset with RDRAM outperforms SDRAM platforms? Then please consider that the i840 uses dual channel; using a serverworks server set III chipset with dual memory banks, a Tyan Thunder 2500 is very much faster than an Intel Outrigger OR840 motherboard. So performancewise RDRAM cannot show any muscle: for every RDRAM board you show me, I can show you an equivalent motherboard in SDRAM technology that's faster.
Is it then because RDRAM is a NEW technology, that can grow and replace SDRAM that is OLD technology? Sorry, wrong again. RDRAM goes back to 1990, when Rambus was founded. The current Direct RDRAM is the 3rd version (after Basic RDRAM and Concurrent RDRAM) and still Rambus doesn't outperform its contemporaries; and after 10 years and 3 major revisions you'd expect that they could...
Is it then that the technology used is so ideal for PC's that it will automatically replace SDRAM? Well the present shows exactly the opposite. Even motherboardmakers that used RDRAM-enabled chipsets scrambled to use MRH-S SDRAM translator hubs. And for the near future, the same holds. Only Intel, held by a strangle-contract (Dutch expression) supports Rambus, and by now not even whole-heartedly.
Like I've said before, I only see Rambus RDRAM continue in notebooks where power consumption is a factor, and in small devices where granularity is important. Rambus holds no real bandwidth advantage, and imposes quite a latency penalty; also the maximum memory size is very limited: 512 MB per channel vs. 4 GB per bank for a Serverworks Server Set III WS chipset (the one for workstations; the server version naturally goes a lot further)
"I'm a stream of noughts and crosses in your R.A.M."