Considering my entire point was. Dual GPUS are where the memory gains are most prominent. I really dont see how you can say that comparing memory usage in SLI configurations to be bad tests. They arent. The 9800GTX + failing at 1680x1050 has absolutely nothing to do with its ability to scale. This can be seen in single "AND" multi GPU enviroments.
My point is, comparing different cards doesn't validate your memory claims. The cards are different hardware, and can be faster/slower depending on a myriad of issues. You should already know that.
Take a card that is offered in 512/1GB configs, and SLI them. Compare those. The same card. Then you'll have something valid. Until then, it's just benchmarks of different cards, when the memory configuration is just one of many variables.