Maybe Bart can explain what's going on here. ... The graphs for the D7000 on DXOMark are very linear and straight. If you check the values for DR and SNR at 18% at each ISO setting, you'll see there is approximately a 3dB fall in SNR that corresponds with each fall in ISO, and a fall of approximately 1 EV of DR for each fall in ISO.
I suspect that the figure of 6dB that is sometimes equated to a change of 1 stop of exposure applies only to the deep shadows. In other words, at very low signal levels a change of 1 EV is equivalent to a change of 6dB in SNR.
I'm no Bart, but we must distinguish two things here:
a) "DR" in the sense of total SNR; the rest of maximum possible signal at a given Exposure Index ["ISO"] setting (limited by highlight clipping) down to the noise floor
b) SNR at a given level of signal, like the level that gives 18% of maximum level, which is then the ratio of that signal to the noise level in parts of the image at that signal level.
Putting aside the fact that I think sense (a) is a bit silly, since it is mostly measuring amplifier clipping of highlights due to excessive amplification, and could be avoided in a truly "ISO-less" sensor by applying that same analog amplification at all EI settings and then adjusting in the digital domain, here goes:
In sense (a), each doubling of ISO setting corresponds to a halving of the maximum photosite output level before clipping, while leaving the noise floor unchanged, so reducing the SNR by a fact of 2, or 6dB.
For the 18% tone level comparison (b) doubling the IS again halving the signal level (photoelectron count) corresponding to that 18% level, but at all but very high EI settings, the noise at 18% level is dominated by shot noise, not sensor dark noise, and halving the signal also reduces this shot noise by a factor of sqrt(2), so that the SNR, being roughly "mid-tone signal to mid-tone shot noise" is reduced by only a factor of sqrt(2), which on the logarithmic dB scale is 3dB instead of 6dB.
By the way, I would expect that as you push to extremely high EI settings, the increase in 18% SNR for each doubling of EI gradually increases from 3dB towards 6dB, a sign that sensor dark noise has become significant even in the "18% mid-tone" region. Is that what you see?