When large JSON structures are being stored but don't need to be queryable it may be more performant to handle the compression at the field-level as opposed to the network level (similar to what is described in https://medium.com/idealo-tech-blog/advanced-mongodb-performance-tuning-2ddcd01a27d2).

This would isolate compression and decompression on the client side (unless network compression was used as well), however the compressed field would be persisted as-is (compressed) to the cluster.