The scope delievers exactly the same amount of data in HighRes mode like in normal mode. So there are no extra samples I could average. And why should I select WORD mode if I have to throw away 50% of the data and average the rest on my own?
The Rigol implements HighRes averaging
between sample memory and display memory. Both are 8-bit memory, so once the averaging is done internally, the result is down-sampled to 8 bits again. If you want 12 bits (or anything >8 bits), you will have to read the sample memory and do the averaging in Labview. If you read the display memory, you're reading the already down-sampled high resolution waveform.
But just in case you're not aware of it, High Resolution mode (i.e. successive-sample averaging) produces the effects of a low-pass filter (or multiple notch filter):
...which equates to the following bandwidths when running on the DS2000: