I think scope can still achieve decent resolution since they end up doing oversampling all the time. So the only weak part is when it comes to none repetitive signals where you can't average things, and the voltage refs used aren't as good compared to proper meters.
Personally I would say +-0.5% within a Fluke is pretty darn good for every day work and from psycho0815's test even the cheap ass Rigol aced that.
Hypernova is very right.
The dc accuracy of the scope is normally quoted for single point measurements which are 8 bit and only around 2% accurate. But the OP's question was regarding measuring dc voltage like a voltmeter so we can use averaging.
With 500 points on the screen and 256 point averaging on the WaveJet (for example), assuming the presence of some white noise, the number of effective bits is much more than 8, if the noise was white and evenly distributed then the number of bits would increase by (9+8)/2 giving a total of nearer 16 bits though in reality the noise floor of the scope would limit this. Also the error is decreased significantly.
I set my WaveJet for 256 point averaging and average reading of the dc wave form and set my power supply for as close to 1V as possible measured on the bench multimeter.
The bench multimeter reads 1000.45 mV and the average reading on the scope is 999.2 mV so the difference is only about 0.1% - much better than the single point accuracy of 2%.
If you're allowed to use saved waveforms then the averaging could take place over 500,000 or 1M points. Of course the scope's linearity and voltage references presumably become the limiting factor but on the WaveJet at least these seem pretty good.
The above measurements were made after not much warm up. Now the meter is reading 1000.07 mV and the scope is reading 1.000V. Perhaps scopes can make good dc measurements.