I suspect that it works like the leafnode Usenet server did.
A full Usenet feed is a lot of traffic.
Leafnode would only download or update a newsgroup's contents when first requested by a client. But once it did so, it would store that data and make it available to other clients. It kept bandwidth requirements reasonable for Usenet servers with a small number of users.
The idea here is presumably aimed at scaling -- to basically try to only download what your users want, but once it comes down for one, to let all the others use it. Optimizes for your instance's bandwidth.