Think about the physics for a moment, the photon flux at the sensor limits the potential signal to noise ratio based on the number of levels of quantization desired, the size of the pixel, the exposure/sampling time, teh well saturation limit, and the quantum efficiency of the sensor. Quantum efficiency of back lit sensors can approach 90% but is more typically about 70% and front lit sensors are about 20%. A 20% Qe sensor effectively throws away 80% of the light falling on it. When sensors are "pushed" they really don't get more sensitive, they merely crank in more amplification (a combination of digital by shifting and analog by increasing gain). The base sensitivity is determined by the flux density and the number of photons necessary to saturate the sensor cell. So as you peer at your images, looking for the evidence of noise, just do it with the knowledge that each time iso sensitivity is doubled, you lose a bit of quantization level. A 100 iso base sensor that produces 12 "real" bits of information that is pushed to iso 16400 has only 6 bits left. The result is banding, maybe you don't see it because noise tends to obscure it, but it is there and can be measured.
-bob