This sort of thing is why containers without a full OS underneath are probably the wave of the future. You don't need, or want, 95% of the stuff that typically runs in the background on Linux.

Is systemd now able to work within user namespaces?

Solaris introduced zones (like containers in linux world) like in 2003. zonectl, zonecfg, etc are some tools used in the solaris world to manage themm.

Silimarly, FreeBSD has had 'jails' for a long time as well, and NetBSD has more recently come out with 'rump kernels' to fill the minimal OS role.

Rump kernels are something quite different from containers or jails. They provide just enough of the kernel interface to a kernel device or filesystem driver to allow that driver to be linked in with other code to form a user-space process. If this sounds a lot like FUSE to you, you're onto something: NetBSD's user-space filesystem layer, puffs, is based on rump architecture, and there are at least two FUSE API wrappers for puffs.

Where it gets interesting is when application code is linked together with all the rump kernel code the application needs, forming a unikernel. See Antti Kantee, "An Internet-Ready OS From Scratch in a Week", http://blog.netbsd.org/tnf/entry/an_internet_ready_os_from