If RAM optimization is best in 3's, then how would you organize chips on the risers, if there will only be 2 of them?
Here's how it works:
Each riser has 6 slots. Those slots are in pairs. So you have 2 slots per channel. Possibly using alternating colors to distinguish the primary and secondary slots for each channel. (Look to some of the Core i7 boards available now).
All nice and dandy, but what are the gains for someone who's actually going to work with the future Mac Pro ?
If you look at the current Core i7 and how it compares to the Penryn parts, the evidence is clear.
In the case of a server/workstation, the gains should be more evident, given the level of memory access that these systems are designed for. Whether it's used in this fashion, only an individual user can answer.
RAM silly expensive as of now, and there is no substitute for size when it comes to RAM yet, crunch the numbers as much as you wish.
It's the nature of server memory. If additional memory is needed, you'd have to buy more. No different from the current model, as most members have upgraded through 3rd party sources.
However, the costs aren't as bad as the FB-DIMM when it first came out. So though it's not cheap, it will be a little less expensive this time around.
HDD performance will remain to be a bottleneck for years to come , new graphic cards sound better on paper, but might still lack proper drivers, and chances are no affordable GPU for 3D apps will be offered by Apple, as usual.
HDD's are a bottleneck, and have been with single drive solutions. But it's the result of a compromise, as the consumer desktop market is aimed at being cheap.
f you're running a workstation/server, you have a high speed, professional option. RAID. Not cheap, as speed never is, but people/companies pay for it if they actually need it.
Drivers and firmware are always a problem. Nothing is developed to a point it will absolutely work with anything, with absolutely no problems what so ever. It's just a sad truth of computer technology, and it seems to be getting worse IMO. Products are being released too prematurely.
Oh, and Intel is loosing it big time, really hurting right now. Not a good time to invest in new hardware, if I may say so.
R&D is likely to suffer first, along with quality control .
From my POV, it already is. Low margins and greed have a negative effect on R&D. No investments are being made while a current technology is milked for every cent possible. Then that company finds itself in deep **** with nothing to sell.
QC has been dropping for awhile. Failure rates have gotten higher than they should ever be. Seagate's 1TB & 1.5TB drives are a perfect example. Newegg's customer review sections are another hint.

Do the math on DOA products as a minimum.
It's noticeable in other products as well, just look around.
I still don't buy into 100% of the SL hype. There is only so much that benefits from being multithreaded.
It will only be able to do so much, until apps developers manage to write code that can take advantage of the multi threading capabilities. The software tools aren't common/standardized yet, and still being developed last I checked.
Even then, some programs may not scale well. Particularly the simple ones.