He is using a variation of the REDSPEC shaders which is a sophisticated SSS shader.

The samples/pixel comparision is a very tricky one. Let me give you an example.

If you take a V4 with a complex shader (like v4) I get around 400-500 samples/pixel in a scene like that (720p, no closeup), 30 seconds, but in a video I probably need less.

If I were to put her in a glass bottle it would drop down to 200-300 samples/seconds and I would need more samples and I would probably need around 500 if I turn on hotpixel removal so that would make it a minute

If I would use PMC instead of path tracing I would get around 70 samples/second but I might need only 100 to get a good result with the glass bottle (with some caustic blur added)

If I were doing a closeup, I will get lower number of samples, but I could get by with a lower number.

What I found is that the number of samples needed really depends on the amount of available light (more is faster) and how much light needs to travel through specular materials. A light in a glass bulb will need a lot more samples as one without the bulb.

These estimated times are for my machine which has a dual titan for gpu rendering, so pretty fast as well. With a single 780 double the amount of time needed.