In relation to the backup procedure described in the online documentation (https://docs.mongodb.org/manual/tutorial/backup-sharded-cluster-with-filesystem-snapshots/), I have the following concerns:

• You could lock a shard indefinitely because of SERVER-14396
• After stopping the balancer, chunk moves still can occur if auto-split option is enabled.
• The data on each shard backup could reflect different points in time. However, there is a warning on the document pointing that.
• The balancer will write the chunk migrations using w:2 write concern. If you have more than one secondary or a hidden member that could be problematic.