I used to have a SparcStation running NetBSD that I used in a way similar to what this paper proposes - testing my software in an environment where some of my (implicit) assumptions were no longer true (a non-Linux system running on a big-endian CPU that caused a bus error on non-aligned memory accesses) to uncover bugs.

So, I like the idea.