There are several factors that will define the rendering speed, in order of importance:

1- How ell is done the software, a poorly done software is very slow.
2- In which language is written the software:

• A scripting language is the slowest thing.
• Next comes C++
• Plain C is much faster than C++
• Assembler language is several times faster than C.
• An optimized assembler program taking into account scheduling, pairing instructions and using SSE extensions can be two or three times faster that normal assembler.
  Note: You needn't do all the program in assembler, only the parts that affects speed.
  3- Now comes the hardware and the most important is the memory speed (FSB and memory specs). For rendering CPU clock is less important than memory speed.
  CPU speed can be important for not optimized software, but for optimized software the limiting factor is memory access just because the amount of memory used is big and doesn't fit in the CPU cache.
  An unordered memory access runs ten or more times slower than the CPU clock.

As for GPUs the result is questionable, is most a sell, buy, buy and spend more $$$.
A well written software that use pure CPU rendering is faster than one that use a video card GPU.
A GPU is a specialized hardware so in theory can perform much faster the rendering, so why is slower that using CPU that is not specialized for the task.
There are two reasons:
1- As I said before the limiting factor is memory speed. A GPU can render faster than the CPU, but it needs data and the data must be prepared, read and transfered to the GPU, so the gain is lost with too much memory access.
2- DirectX. You need DirectX to talk with the GPU and DirectX is a slow software.

The advantages of using video cards, specially with games is:
1- Programmers with little knowledge can make games.
2- As the CPU has very little to do while the rendering is done by the GPU you can use the iddle CPU time for the game strategies.

Stupidity also evolves!