I consider them stable if they'll run MemTest overnight, and Orthos or SP2004 for 12 hours, using the blend test.
edit: With no errors in MemTest, of course.
edit #2: Don't run Orthos/SP2004 for 12 hours, before you decide to take your overclock higher. Just do 12 hours for the final run, after you've decided on where you're going to keep the overclock. ~1 hour is fine, as you work your way up the Mhz ladder. Also, it isn't really stable until you've used every app that you're going to be using, and had no problems. Orthos or SP2004 isn't the be-all, end-all in stability.