I am not fully aware how IPFS operates, but wouldn't it at least solve the back-end mirroring? Front-end servers would then "only" need to access IPFS for continuous syncing of metadata (for search) and fetching user-requested files (upon request).