I don't know how long it takes for the typical chipset verification (and subsequently, most motherboard design, and verification with the chipset), but could this simply be an economic move?

If there were enough time between the introduction of the P3 w/i820, and Intel had enough time to redesign the i850 to be merely dual channel like the i840, then it might have been simply a political/face-saving view.

People hate rambus as a technology (often due to flamers), and even more hate rambus as a company. If (this is the if that I don't know) there were enough time to redesign the i850 to use only two channels, then they'd be able to do a few things:

A) keep the cost of the system to sane levels - Intel must have known shortly after the i820 that RDRAM might not be the "next big thing" (right away - it still has niche markets for sure, and may expand in the future), and might not become commonplace.

B) They really don't seem to need it. Of course more is better, but perhaps this was an internal move as well? While yes, the Itanium has better SPECfp scores, with increased bandwidth, might the P4 have scored too high? I don't know how the politics work within Intel, but it would be rather embarassing if Intel's "miserly anchient x86" architecture could beat their brand spanking new architecture.

Intel has made some slip-ups, for sure, but the additional costs associated with more RIMM's might detract from Intel's prestige more than they could take. Remember, the P4 is a desktop processor, not a backend server

Besides, ServerWorks is working on a 4-way interleaved DDR solution for 4-way SMP (one chipset). I think that a uniprocessor in such a system could score rather nicely in FP/bandwidth intensive benchmarks.