The quantization steps in the displayed noise signal make me assume that the scope's "true" capability is just 100 mV/div with roughly +/- 1 bits of noise (which may be well acceptable), while the settings below, down to 10 mV/div, are likely just implemented by digital amplification/zoom of the signal captured at 100 mV/div (w/o spending any extra ADC resolution and low-noise pre-amps). So IMO this does per se not look like a bug, but rather a (cost-driven) design limitation of this model (at the bottom line, my 6074BD seems to be significantly better in this regard - just wondering btw, what's actually the price of the IDSO1070?).
I'd consider the difference betwen the two channels within tolerances (it's obviously not factor 10). There is also a small noticeable DC offset on channel 2 (maybe you get rid of it when invoking "calibrate", but I would not be sure).