Prior to incremental WT backups, MongoDB would copy all of the filenames from WT into memory and return them over the wire as needed. This wasn't particularly costly because internally, WT would already keep all filenames in memory.

Now with incremental backups, WT streams the filename/block information from an underlying table. There would be gains (in terms of reduced working set memory) if MongoDB advanced the WT cursor as the external request needs more documents.