I initially chose GTA IV based on the fact that it had a built-in benchmark tool, and based on the fact that it was considered a system killer at max settings. Realistically, I don't believe the GTA IV benchmark tool offers a good representation of the actual gameplay experience. The benchmark is very much a "corridor" run on a dense city street, while the game itself is open world and many locations offer variety architecture and/or a distant horizon. Additionally, the benchmark is set during a nighttime setting. While this is potentially good for showing off the game's neon lighting effects, it offers no sun, clouds, atmospheric coloring, lens flare or texture variety in the sky.
Additionally, I found that the game hit a CPU limit of 45/46fps at 1680x1050 on a Radeon HD 5830 - not exactly cutting edge hardware. The one benefit of the game is that the graphics option screen offers a reading of your system VRAM and calculates how much is needed based on your selections. By default it won't let you go over the limit of your video card.
With a couple of command line switches, the game allows you to max everything out well and beyond what your system is "capable of." Down side it that overloading the video card seems to have little effect. I'm not certain if the non-existent limit is due to the horsepower of the Radeon 5000 series, or limitations of the benchmark tool.
The only time I found the limit to be an actual impediment is when I tried to max out the settings on a the Radeon 56/44/54xx series with 512MB of RAM. At that low of a framebuffer, the benchmark wouldn't load. Otherwise, I could overload the 1GB VRAM all I wanted, and the game performed reasonably well. Considering the issue of overloading the VRAM, the addition of the second GB of VRAM didn't effect performance.
You can hit 30fps with everything set to medium. Performance never moves about that mark.