Discussion of ``The Dantzig selector: Statistical estimation when ppp is much larger than nnn'' [math/0506081]