It would be useful for verifying that the crash really was controlled (cases like bug 119963). To improve testing performance it could also avoid actually crashing.
Very neat idea! We could also consider avoiding costly generation of crash logs for regular crashes for tests with [Crash] expectations.