I didn't get too much into the original article but my impression is that he is using Scot encoding (essentially a 1-to1 translation of pattern matching) instead of church encoding (somethign equivalent to folds).

Scott encoding doesn't get much publicity but its perfectly valid and much more intuitive, IMO.