A blend loop doesn't complete until ~12 hours. You need to test every FFT if you want to ensure stability. That's really the bottom line, you can pass 4 hours of P95, but as soon as you hit an unstable FFT it could crash immediately. Which results in some games having issues, where other games won't.
12 hours of Prime95, always using the latest version (27.7 B2 currently?), with both error checkings and max memory set to 80% of your current memory. This was harder to pass than 100 passes of IBT/LinX on their highest settings on my 3770k. Prime 27.7 is pretty intense now, I would consider it all around superior to LinX/IBT with it's new AVX FFTs. However, I still test with both just to make sure.
So test every FFT at 15 minutes each = 17.5 hours? That's 99.9% stability?