When a sharded cluster has a large number of shards, it becomes more likely that one of those shards may be experiencing a problem at any given time. When this occurs, scatter/gather queries break for the entire cluster.

It would be better if MongoDB could ignore problematic shards and return the results from the rest of the cluster in such cases. This should not be the default behavior, but it should be configurable on a per-request basis for those requests that can handle incomplete results.