Remember that with reusability in play, the guidance system can decide "screw the first stage landing, the payload is more important" and simply switch to an expendable launch profile if it turns out that you have more engine problems on the first stage than you designed for. Then you use the reserve fuel to get the $300M satellite safely to orbit instead of getting the $30M stage safely to the ground. The customer is happy and you've only lost a unit for which you have a dozen of direct substitutes in storage anyway (SpaceX already doesn't have enough storage space, apparently!).

It might depend where in the flight the engine fails, assuming we're talking about the first stage. If I had to guess, a GTO mission (placing a payload in a 36,000 km x 350 km elliptical orbit) could still be successful with an engine failure in the first stage after 65-75% of the fuel had been burned, and the vehicle was generally lighter, allowing the burn time to be adjusted longer to reach the same velocity at the point of first stage cutoff and separation to ignition of the second stage. But this guess comes from dozens of hours spent playing Kerbal Space Program....