My first example is the Simulanalog Guitar Suite. They've got their tone down to less than -40dB of difference to the classic original stompboxes and amps (mathematically provable). And that's on test tones, which as we all know always accentuate any audible problems tenfold - after all, that's the point of them.

My other case in point is Guitar Rig 2, and this is where I insult all the 'golden ears' bearded fellows. I would say any lack of realism that can be found using GR2 is down to a bad patch programmer. The tones found in GR2s devices are audibly very close to DIed real-world equipment - most of the problems arrive at the cabinets-and-mics stage, that is, the very beginning and end. People choose inappropriate mics, cabinets, and cabinet sizes, and stack them up when they don't need to. Suddenly they've made a mess of a serviceable chain.
Of course it pays to know exactly what you're doing with the sound from the start, and experience with the devices help too. Ultimately it's the entire rig that processes the sound so if you chain the devices badly, or worse use the wrong devices altogether, you're going to get ugly guitars. Crap in, crap out.

As a little test, I'd like to do a double-blind test with some people that say they can hear the difference between a well-written GR2 patch (with a DIed guitar) and an identical hardware chain.