[ijlx]

Per https://wiki.archlinux.org/title/DeveloperWiki:NewMirrors a full mirror of the arch repos would be over 50 GiB. AUR would be impossible to tell, as you have to grab sources and build it yourself. I'd imagine fairly massive though.

I can only really speak for arch as that's what I'm familiar with, but I'd imagine the mirrors would not be happy about all the wasted bandwidth. The whole idea sort of seems unnecessary to me.

[ghostpepper]

What about pulling in binary packages the first time you try to run them. Or pulling in shared library packages the first time an app tried to load them.