when creating a neighborhood based recommender,there are various distance-measures available (e.g. in Apache Mahout) and I can find quite some practical examples for most of them. But I never stumbled over examples where the Manhattan distance (a.k.a city block distance) outperformed others.
I'd be happy to find some ;)
Manhattan distance is rarely useful; it's there mostly for completeness. It might be useful in cases where your ratings are really discrete values like counts, and it has some relation to edit distance, so might be useful if the items were... strings. But that's probably not a real use case for recommenders.