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