I'm assuming that you've already figured this out given that it's been quite a while, but it's an easy question to answer so I figured I'd leave the answer here for posterity.  In the sequential case, customize a BFS visitor to throw an exception on the first path encountered longer than n.  In the parallel case you could either add an all_reduce per epoch to the algorithm explicitly or customize the queue to embed the 'terminate_early?' predicate into the distributed queue.  This is (IMHO) the canonical way to handle depth-limited searches in BGL and a method that I find myself using repeatedly.  It's so trivial to implement (at least in the sequential case) that I've never felt that it warranted a separate algorithm.

Regards,

Nick