The problem is probably your OS. I have never been able to effectively use a card when I have less usable system memory than the amount of video memory on the card (per GPU). For example, to use a 280X with 3GB of VRAM, I need 4GB of system RAM and a 64-bit OS. The problem does not scale with more cards, so I can run 3x 2GB cards on a 32-bit OS, but have never been able to run even a single 3GB card. I have not been able to figure out why (the miner only actually uses about 50MB of system RAM), but that's just the way it works.

The "thread-concurrency" setting is limited by video and system memory, among other things. A 2GB card likes a TC of 8000 to 16000, a 3GB card 21000 to 24000, and a 4gb card 30000 or higher, but with a 32 bit OS you probably won't be able to run a TC over 16000.