I don’t understand your point - for clever dictionary-based crackers the randomness of the selection process is not important as the software uses strings instead of characters to cobble together a password guess; I thought the description in the Ars Technica article (linked in Schneier’s post) was a very readable one.
I remember a quote of some security researcher (?) along the line “finally an answer to the staplehorse-thingy”, but don’t find it anymore.
