If you are interested in implementing the power crust algorithm, or its predecessor, the Crust, there are some things to watch out for.First, you'll need a good program to compute the Delaunay triangulation, the dual of the Voronoi diagram. We used a C program called hull, by Ken Clarkson at Bell Labs. Hull uses a lot of memory but it is very robust Do use the -m and and -r flags!!! We had to hack the source a little to get it to compute power diagrams, and to figure out how to read the data structure it produced instead of regenerating it from the text output.
The CGAL library has a robust Delaunay triangulation program, equally slow in our experience but possibly using less memory, and with a good interface for power diagrams and access to the data structure. CGAL is a C++ library with a lot of other functionality as well. They say they are releasing a much better Delaunay triangulation program sometime real soon now; check their Web page. If it's out, it's probably your best bet (look for something after 2.2, the current release). I think you should randomly shuffle your input points before using CGAL's Delaunay triangulation. There are some other very good codes in development; stay tuned.
The second thing you need is a robust way to compute the locations of the poles and other Voronoi vertices, and the radii of the Voronoi balls. This computation is numerically unstable for very flat tetrahedra, and if the polar balls are wrong then the output is wrong. We had vary good luck using a modified (3D and then 3D power diagram) version of this code offered by Jonathan Shewchuk, and using his robust determinant functions. We found that if your input points don't have too much precision, either the position of the Voronoi vertex is fine or the determinant in the denominator of the computation (see the code, above) is exactly zero (probably someone really smart could prove this...). We rounded the locations of the poles to get this nice behavior in the power diagram computation as well.
You'll definately want to read our papers before getting started (especially the power crust paper from the Solid Modeling conference). And you might want to try out our software!