I get it, for many this is just a marketing move. I respect this opinion but I now see this differently. I recently purchased a GPU based rendering software only works on Nvidia CUDA programing language on gamonig or Pro cards (OctaneRender). For this software the memory installed on the GPU is critical as this is the only memory the software uses. Two 2 GB 460 card are more appealing to me as I could use a combined none SLI 2 x 460 card setup for a total 670 CUDA cores and 2 GB of total memory for less than $600. The software is capable to use the combined CUDA cores power of 2 GPUs but the memory needs to load all resources on each GPU so It wont see a 4GB of total memory.
Also, I read some researchers using CUDA could use the 2GB memory on a gaming card as a cheap replacement for 4GB TESLA GPUs. I think this move is appealing to none gamers, but I can't say some gamers wont find this useful.