I just had a look at all the channels on mine. I thought of a couple more tweaks as well.
a) Set "High res"mode
b) Turn on the low pass filter
c) Turn off all the lights
d) Push the vertical scale knob. This allows you to fine tune the vertical scale (I set it to 400mV/div for the readings instead of 500mV).
e) Attach the ground clip of the probe to the ground point underneath the calibration signal output. This seems to reduce the noise a tiny bit.
With these settings I got the following Vpp readings on my DS1054Z.
CH1: 3.02
CH2: 2.99
CH3: 3.05
CH4: 3.05
For comparison, in 'normal' mode with a 1Vpp vertical scale I get these readings:
CH1: 3.08
CH2: 3.12
CH3: 3.12
CH4: 3.04
The values weren't stable though, those are just the numbers I seemed to see most. All of them wobbled around and quite often went over 3.2V (sometimes even over 3.3V).
I also noticed that when the numbers wobbled they always went up/down in steps of 4 (ie. values were 3.04, 3.08, 3.12, 3.16, 3.20, etc. with nothing in between). The resolution of the ADC must be 0.04V at that vertical scale - that's more than 1% of the reading!
Conclusion: DSOs aren't multimeters, they simply don't have the noise immunity or number of bits in the ADC.
(or at least the DS1054Z doesn't... I'm sure the old green screen HP27634837684A will)