Re-balancing shard allocation

This is strange then. There must be quite a difference between the numbers of shards on each node. Please could you provide the output of GET _cat/shards?v for us to see the details?