The noise with the fake ISOs *is* linear related to the ISO from which they are derived. There is no point of researching the noise characteristics of those ISOs, and shooting with such ISOs in raw is not only useless but counterproductive.
See List of read noise per ISO
Interesting, but your measurements are only about means and deviations which are quite a reduction of noise distribution.
Actual distribution measurements show that, for example, the curves at 12800 and 25600 for the 5D Mark II don't have exactly the same shape, and that the differences depend on the shutter speed.
This may be the result of a variation in the readings (the sensor temperature might have changed), but can also be the result of a more subtle way of handling the problem of the so called "fake ISOs".
Of course I agree that the "ISO H1 and H2 extensions" (to use the Canon words) are probably mainly generated by multiplicating the values of ISO 6400, while regular ISO are made by using different gain in the analog/digital conversion process. However, displaying the curves is a way to check this, and doesn't cost lot of time.
Shooting in real life with "fake ISO" in RAW is a convenient way to avoid the problem of having to think about under-exposing when shooting and then later on boost the levels in post-processing... Of course, some highlights might get lost in the process.