It's a nice method to have in one's tool kit, and thanks for posting the tutorial.
However, IMHO, you have given it an incorrect and possibly misleading name. The technique you appearing to be sugesting is nothing more than generating a classic luminosity mask adjusted so that the most contrast in the mask occurs in the tonal range between the hair and its background.
Also, I see absolutely no need to suggest that this technique needs the full dynamic range of a raw file. In fact, if the part of the image in question is bright enough so that one can visually distinguish the hair from the background in a temporary, brightened version of the processed image, one certainly can use the image itself to create a suitable mask. Of course if the image happens to be 16 bits/channel, it will provide somewhat smoother transitions than an 8bpc image, but one doesn't need to suggest that the full dynamic range of raw data is needed.
Perhaps a better name for your tutorial might be "Using luminosity masks to extract complex objects from their backgrounds".
With respect to its utility, the major problem in general applicability of this technique is that if you set the point of maximum contrast for best separation of (say) hair from background in a darker part of the image, this will not be the best setting to extract hair from background in a better lit portion of the image.
What one needs to overcome this limitation is either:
(a) an adaptive technique such as used in some commercial masking / cutout software; or,
(b) temporarily but dramatically reducing the contrast of the image at the lowest spatial frequencies while retaining (or even enhancing) contrast at high spatial frequencies. This can be done, at least to some extent, in ACR by decreasing the clarity slider and compensating for this by increasing the sharpening sliders. The better way to do this is by using either do-it-yourself or commercial frequency separation techniques such as Topaz Detail.
In fact, the commercial masking software packages use several different algorithms to initially determine boundaries and continue existing boundaries. For example, in some areas, they may find that a luminosity based technique (such as discussed above) is best, whereas in other areas, they may determine that a hue-based discrimination technique is better and switch to that. Also, almost all of them are adaptive in the sense that they use local image properties to segment the image. For example, as they wind their way around a head trying to determine the best separation of hair from its background, they will automatically recognize where the background brightness or color changes and adapt to that.
HTH,
Tom M