The DS1000Z series only has 2 analogue ranges, with an attenuator switched in for 500mV/div up to the max of 10V/div.
All other gain control is done digitally inside the HMCAD1511 ADC (it has internal precision >8 bits, and thus can do digital gain control over a fairly wide range), which Rigol has taken advantage of in order to make the analogue front end very inexpensive. However this digital gain has a max of 50x, with missing codes at gains higher than 32x (see HMCAD1511 datasheet).
As Rigol is using the same analogue gain for all ranges 200mV/div and under, best case is that they'll be hitting >32x digital gain (and therefore the ADC output will have missing codes) at 5mV/div, and will be beyond 50x at 2mV/div and 1mV/div, so will need software gain for these two settings. There will also be some decrease in ENOB as digital gain is increased - the datasheet specs 7.9 bits at 1x going down to 7.6 bits at 10x gain (which actually seems like a pretty good performance to me!)
tl;dr: it's already run out of steam at <5mV/div, don't worry about trying to do any better! You won't get amazing performance out of a DS1000Z scope, but you might get amazing value if your use doesn't push it's limits.