Obviously a few people are confused about SLI and the video memory involved...
SLI does not double your effective useable video ram. NVIDIA's implementation of SLI works in two ways: Alternate frame rendering (AFR) or Shared frame rendering (SFR). 3DFX's method of SLI worked in a process called Scan Line Interleave, which is actually where the whole "SLI" acronym came from. Here's how they all work:
AFR: The two video cards "trade off" rendering frames. For example, Card A renders Frame 1, then Card B renders Frame 2, then Card A renders Frame 3, then Card B renders Frame 4. In this mode, both cards use their onboard memory to retain the texture and vertex data for the entire scene, so your total effective video memory is however much is on a single card.
SFR: The two video cards perform a sort of "load balance", where an imaginary line is drawn down the middle of the screen and each card renders it's own "half". This line can move around depending on vertex or shader load, and tries to keep the load semi-equally balanced between the two cards. Because the line could "move" at any point, both cards retain all the texture and vertex data (but may not use it all during every render). Thus, your total effective video ram is however much is on a single card.
SLI: The true meaning of SLI is scan line interleave, where each card renders every-other horizontal line in a rendering. If you were running 1024x768 resolution, that would mean both cards would be rendering at 1024x384, except that one card would be doing the even lines and the other card would be doing the odd lines. Mesh the two together, and you get your true 1024x768 resolution. It's another form of load balancing that isn't nearly as technologically advanced, but was quite effective in it's time. But again, since both cards are rendering the entire scene (albeit at a slightly lower resolution), they both retain the full texture and vertex data of the entire scene. Thus again, your total effective video ram is however much is on a single card.
Two 128mb cards in AFR/SFR/SLI = 128mb of useable video ram.
Two 256mb cards in AFR/SFR/SLI = 256mb of useable video ram.
Two 512mb cards in AFR/SFR/SLI = 512mb of useable video ram.
That's just the way it is, and no amount of driver tweaking or fenagling will help. The only help you get in some circumstances is a slightly smaller framebuffer. In Shared frame rendering or Scan line interleave, each individual video card can use a smaller frame buffer; which could save you a few megabytes depending on resolution. At 1920x1440 at 32bit color, SFR could (in a best case scenario) save you around 10mb of double-buffered (ie vsync disabled) framebuffer space. If you were leaving vsync enabled and were using triple buffering, it could save you as much as 15mb of framebuffer.
But since that's pretty tiny in comparison to the rest of the video memory on a normal card, AND that's also a pretty stinkin' high resolution, it's doubtful the 10-15mb of memory space will gain you much performance. You'll more likely be either CPU bound or fillrate bound at that kind of resolution.