I published a little CUDA app that allows one to render amazing "plancton" like fractals in a voxel representation. It needs a *fast* nVidia chip though - even on my GTX 260 the updates after changing the parameters take half a second or more. The CUDA source code and a Win32 executable are attached inside the ZIP file in this thread.
This program uses a slight generalization of the Mandelbulb idea where instead of using a spherical coordinate system I use two separate rotations based on axes that one can freely choose relative to each other.
With an adjustable exponent, and two extra degrees of freedom for pointing the axes relative to each other endless fun is guaranteed
You can also rotate the plancton, pan around and zoom - but deep zooms are impossible due to the limited resolution of the voxel space.
Also find screenshots and the Win32 binary and CUDA source code here: http://forums.nvidia.com/index.php?s=1bf094526ea25174ffb8815dbbad1df7&showtopic=153257
and here http://www.fractalforums.com/mandelbulb-renderings/a-new-class-of-bulb/
On slower cards you can start VoxelPlancton.exe with the --size=128 or --size=64 parameter (this works even on my laptop with a 9600M GT card) - but the rendering is less detailed then .Arguments can be passed from a command prompt or by creating a short cut to the exe and appending this argument in the target field of the shortcut, but leave a space character inbetween.
It was a fun two-day project and I might explore further to find for a raymarching solution (I need the derivative, but my equations rotate with quaternions based on arbitrary axes - so it is a hard problem for me to crack). The linked "A new class of bulb?" thread in the Mandelbulb Renderings subforum has all the maths details and some C++ source code for the CPU.