I think you have an misconception about how sci-hub works. Most of the content you can access trough sci-hub is not hosted by sci-hub, what happens when you request a fresh new paper is that your request is eventually served by a computer in the network of one of the universities with subscription to the journal which published that paper, using donated or stolen credentials of some high profile scientists if needed. Sci-hub already uses IPFS to cache some papers, it can not move to IPFS network because it is not a library of content like LibGen (which is often confused to) but an actual hub which decides how to serve your request in real time.