> Further, Martinus' solution was the fastest (even with the additional  
> code to normalize the points). In calculating the 20,808 faces for an  
> octahedron of frequency 50, the times on my computer were:
>
> Martinus:  4.32 seconds
>     Gavin:  8.61 seconds
>    Warren: 12.81 seconds
>    Dennis: 80.49 seconds

Woah ;)
I tried to keep the code as simple as possible on the expense of  
calculating the vertices more than once but didn't expect the result to be  
this slow...

> For reasons I couldn't figure out, Dennis' solution returns a bunch of  
> NaN points when dealing with the icosahedron primitive.

The slerp method I used assumes that the incoming vertices are normalized  
which the vertices of the icosahedron in your testdata aren't. (It's  
evident that the length of Vector[  1,  GOLDEN_MEAN, 0 ] is > 1)

-- 
exoticorn/farbrausch