Things change fast, and I'm probably behind the times... I hope I am only slightly wrong with what I say.
For the most part, high end cards a great at rendering in the style of OpenGl or DirectX - scan line conversion, with vertex shaders and all that, and do so very well at high frame rates. High quality photo-real rendering usually involves ray tracing, photon mapping, Metropolis algorithms and other techniques that don't translate well into hardware, and rarely can be done real time (w/o spending big, big bucks). The high end card may pay for itself during the modeling and lighting phase, where the user interactively alters meshes, moves lights, applies textures etc. while the final output render would be done in mostly in software. Typically, a designer spends a lot of time modeling and texturing and adjusting the view, and (at least on a good day) taps the "render" button only once.
The essential stupendous point: be clear, when deciding on a graphics card, on understanding whether it's benefit is for during interactive scene design, or for the final renders.