If you click on the "need more help?" link it explains that the CPAN distributions are laid out along a Hilbert Curve. The resulting shapes are dictated by where along the curve one namespace ends and another begins.