Poor Refresh Performance only in Service

We've had several semantic models go from < 5 minute refreshes to > 1 hour in the last 3 days.   I've tested various things, but it boiled down to an unexplainable result:   I can publish an exac...