Because the criteria for paywalls are publisher-dependent, and can involve reader's country, IP range, referral source, browser history or some quota of free articles available per month. Articles from sources with extremely restrictive paywalls (ft.com, nature.com) just don't get many upvotes, so the problem is self-correcting.

Alternative solution would be to link to a paywall-free blog post rehashing the contents of the article that includes a token link to the original (huffingtonpost.com or businessinsider.com), but HN's policy at this point expresses strong preference for the original.

Usually the 'web' link is sufficient, but the policy on Google allowing traffic may be changing.