Bulk insert requests not balanced across cluster

Are you distributing the bulk requests to all nodes or only to a single one?

That may explain if you do the later.