A GTX 9x0 is an overkill for any SL viewer: a GTX 660 will give you equivalent results unless you got a super-high resolution monitor tied to it...
The bottleneck in your system is likely the CPU: even the fastest CPUs can't feed a GTX 9x0 fast enough to load it at 100% in a software such as the SL viewer, which uses a single-threaded renderer (i.e. the rendering code only uses one CPU core, and the speed will depend almost exclusively on the CPU frequency, not on its number of cores when they are beyond 2).
To try and get the most out of your CPU, you may enable the multi-threaded optimizations in the NVIDIA driver (this is done automatically by the Cool VL Viewer, under Linux, thanks to an "export __GL_THREADED_OPTIMIZATIONS=1" in the cool_vl_viewer wrapper script, but this must be configured at the driver level under Windows).
That's because the main loop code was optimized the most, giving more time for the CPU to execute the rendering code. It's important to optimize the code where the speed bottleneck actually is...
I don't know if you overclocked your CPU (it will keep reporting its default factory speed in /proc/cpuinfo under Linux, so what is displayed in the About box is not relevant), but if you didn't, your best bet is to try and do it (my Core-i5 2500K is overclocked at 4.6GHz, and it does make quite a difference when compared to the 3.3GHz stock speed).