Technically not difficult, but would add quite a bit of complexity to HN in order to fetch and store at least a fuzzy hash of the article bodies.

From there it's mostly a case for a bayesian classifier (LSI) and some elbow grease.

FWIW, the "similar articles" footer on most blogs is implemented that way.