Brad,

> I believe that L'Ecuyer's generator would make a better choice for
> a reference implementation.

As I share your doubts, I have included the generator you mention
in the reference implementation.

More serious than that: You comment has raised a larger number
of additional issues for me. Mostly about floating point implementations
of the generators (which can be must faster than integer impl.)

I will come back to this later.

Sebastian.