I’ve always been curious what the most optimal “network-portable mountable filesystem image” format for Linux would be (for e.g. being a zero-install auto-fetched-from-the-network “mount and run” format for applications ala Flatpak/Snap; or being a NeXT-alike distribution container for “bundles” — where, as in macOS, it’s just as viable to open the app/document bundle [read-only] from the image, as it is to copy it out of the image.)
Such a format would have to balance the considerations of download time, storage size at rest, mount time (incl. any integrity checking), and access-time overhead when mounted. It would especially have to worry about per-file-open overhead, given that one of the primary use-cases would be rsync-ing the file tree out of the image onto your disk.
It sounds like you’re saying that a SquashFS filesystem image would actually be bad for this use-case, vs. just downloading a regular (sparse) loopback image of an ext4 filesystem over a compressed transport. Which is kind of surprising to me! I would think that the use-cases stated above would have been one of the design goals.
I’d also be curious about the relative merits of the “pure userspace” option — i.e. rather than mounting a disk-image and then accessing the files inside it through regular VFS inodes, instead having your zero-install daemon or whatever just use regular tarballs, by linking in a library like libarchive and using it to get internal virtual handles on the files. (Lots of games seem to choose this option, so it must have pretty good perf.)