Right now it seems as if its up to nVidia to push SLI, while I would most likely be one to use it as an upgrade path, such a thing is precisely what a company like nVidia wouldn't necessarily want. If SLI does increase performance anywhere from 50-100%, then that could prove to be too competitive for any future flagships unless SLI is practically made too expensive to be realistic (ie the motherboards might cost an arm and a leg if not more).
Some people always upgrade to the fastest there is, even if that means spending $500 a year on a video card. Now with SLI you could spend $500 one year, and then $200 or so to upgrade and double up your performance instead of paying $500 for the new part. Now unless the new part packs a ton of features you absolutley have to have, then there's not much reason you wouldn't do SLI unless there is some downfall.
SLI just seems too good to be true for your average consumer although I sure as hell would like to see it but as it is I don't see it happening with ease. I mean as it is two $200 MSRP 6600GTs would most likely easily best the $400 MSRP 6800GT, so why go with the single card solution if you have $400 to spend unless you're planning on upgrading in the future? You could always go with the 6600GTs and wait for the 6800GTs to come down to $400 for the pair.
Alienware's solution seems like a sure thing in the sense that it will actually come to market, but at the same time their 800W system will probably cost a good $5000-6000 if not more which would make sense. I guess nVidia's SLI will depend upon the price of dual 16x PCI-e motherboards as well as the SLI connector that might be sold separately, who knows, they might charge a ton of money for that "simple" connector.