This is a great idea, but unfortunately many runbooks can only be meaningfully evaluated in a situation in which they are useful. That is - you need to instigate an outage (ideally, in a controlled reversible way[0]) before you can "test" a runbook.
If you're following Google's SRE all of your infrastructure should be pretty easy to replicate with a sample of production data. You can try to crash it in calorimetry to your heart's content, away from production systems. You might even learn (or teach) some things through the process of creating and attempting to destroy a test bed of this nature. They (https://sre.google) cover this in chapter 17.
[0] https://sre.google/