Given that current master won't be released until 1.32 in January, after Fedora 27 goes EOL, I'd be OK with dropping LLVM 5 now. I don't need to build beta/nightlies for the distro.
We don't track every rust-llvm patch in our system LLVM. Some have relatively low impact, and some aren't even relevant on Linux. And frankly, a lot of times it's reactive -- we haven't applied your example JumpThreading patch yet, because we haven't seen that issue. (I guess because rustc also worked around it.)
But yes, I do see this as the job of a distro maintainer. I recognize that it may be harder for those who volunteer vs. paid roles, but we can also collaborate on this.
We can and should work together on this. I'm happy to attempt backports myself, and I also have teammates to consult who work on LLVM directly. Your particular example for LLVM 5 may be irrelevant soon, but the offer stands in general.
(edit to add) All that said, I do know the overall problem is still not trivial, but I think it's manageable.