The CTE method of unrolling a hierarchy uses the adjacency model. It's just another way to handle that, rather than cursors/while loops. I've tested cursors, while loops, and CTEs for complex adjacency hierarchies, and CTEs are faster in all of my tests (up to 7,000 nodes on the hierarchy, up to 50 levels). Cursors are the slowest. A semi-set while loop is in between.

Nested sets are much faster to unroll. And they require one table scan per query, instead of a number of table scans equal to the number of nodes in the the hierarchy (cursor or CTE) or a number of scans equal to the number of levels (while loop). The problem with them is if the data changes much, they are much more difficult to handle. (Nested sets hierarchies select very, very fast, but update/insert/delete much more slowly; adjacency selects more slowly, but updates/inserts/deletes very efficiently.)