This gives me the same result as the recursive version, minus the level column. I am porting an application from Oracle, there is a fairly large table that is accessed by "connect by". Rewriting it as a recursive join is not a problem, but the optimizer doesn't really use the indexes.
You're still using a 14 row table, though. Postgres isn't going to be stupid enough to use an index in this case when the seq scan is clearly faster unless you go out of your way to absolutely force it to do so. If the table is going to be "fairly large", that's the size you need to be testing and tuning with.