Microservices and Database Per Service: When is it Okay to Cheat?

We all know the 'database per service' rule for microservices. Everyone says it's best practice for loose coupling and independent deployments. But how many of us have actually broken this rule in production? I mean, for simplicity or a tight deadline, two or maybe even three services might share the same database. Maybe they even use different schemas inside that one instance. What are your stories, good or bad, about going against this idea? Are there times or situations where you found this 'controlled cheating' was okay, or even helpful, for a little while without causing a lot of problems later?