In the early days of the VSL we had all our demos mastered especially for the MP3 format. Nowadays they are mostly simple conversions from a PCM mix file. In my opinion, the bitrate of 256 kBit offers the best achievable compromise between filesize and audio quality. Whenever I prepare demos for the VSL website, I choose this setting, but I can't say for sure that all available demos follow this standard. As a rule of thumb, the more ambient and "transparent" a piece of music is, and the more stereo (read: out-of-phase) information it contains, the higher the bitrate must be to leave the acoustic impression intact.
Regarding dither: Technically, dithering is necessary whenever you change the volume (and thus the bitdepth) of a digital audio signal. In principle dither is just noise which is used to mask the otherwise "grainy" structure of digital auido at lowes volumes. So changing the format from your linear PCM 44.1/24 bit mix to a 16 bit MP3 means that dither should be applied right before the bitdepth reduction (post-fade, of course). It's up to you whether you use UV22HR (a good choice) or another kind of dither like Waves IDR or PowR. - White noise at -96dBfs would do the trick, too, but it will be more obvious due to the missing noise shaping.
Taking into account that MP3-encoding uses a lot of DSP processing, it is always a good idea to leave more absolute headroom than you would in case of linear PCM audio. Especially with brickwall limited material the processing will most likely create intersample peaks that could overshoot (cheap) D/A converters on the analog side, or do other ugly things during decoding. - This is one of the reasons why many MP3s sound so awful, BTW. ;-) - A headroom around -0.5 or even -1 dBfs should give the encoding ample space for its processing.
Using Cubase (or Nuendo in my case), I use the "High Quality" setting just like you. The manual doesn't offer any details, but I assume that this setting uses a slower encoding algorithm which allows for better psycho-acoustical masking of the data reduction errors. It is not the same as the choice a different bitrate (i.e. you could export a 128 kBit MP3 in "High Quality", too). - Other encoders allow for variable bitrates, but it depends a lot on the structure of the signal what works best in each case.
HTH,
