Being familiar with both the nested set model [specifically Celko's push stack] and the CTE model, they each have their advantages and are relatively easy to set up. I recommend the adjacency list for OLTP systems with a sproc to generate the nested set as hierarchical data is needed (it's a cheap operation).

Another option that I feel should at least be mentioned is denormalizing your data, although it limits the number of levels you can have in your hierarchy. Denormalization, however, would be the fastest hybrid OLTP/OLAP solution (nested sets IMO are functional OLAP).

- James