The card does have its own RAM but the CPU has to talk to this RAM so it has to take up space from that 4GB of address space. If you only had 2GB of RAM then the graphics memory would be mapped into non RAM space (but still below the 4GB limit) but as you have the max RAM then there is some overlap, in your case its the worst overlap where all the graphics RAM has to be mapped over the top of system RAM.
Also remember, the Graphics RAM is only part of the address space needed by the graphics card, the GPU also needs some address space as well so it can be programed, it all adds up.
The amount of RAM mapped over does depend on the BIOS as well. Using the same components but a different MOBO I could see 2.9GB compaired to 3.2GB of ram on my old Vista 32 install using the older mobo.
Either way, to see all your RAM is going to involve going for a 64bit O/S no matter what you do.