The first would be learning/demo purposes where the data you store should essentially be ephemeral (imagine training sessions, interactive online tutorials, etc). In these cases it's more about experimenting with the query language and data model than anything else. RethinkDB has used Docker for this purpose to great success.
The other would involve mounting an external volume in order to persist the data and would allow you to isolate and limit the process which would otherwise monopolize the available resources of system (usually memory in MongoDB's case).
I assume since containers ephemeral, data is saved somewhere on the host.