After getting GPO set up with a Gina3G sound card, studio monitors, and a new powerful computer, I decided to try it out by using Sibelius to play back an orchestral piece that I have heard performed by three different orchestras. Having spent hundreds of hours playing in an orchestra and having heard several hours of my music played by orchestras, I thought I would offer some practical advice concerning dynamics.
I have seen some discussion here of arriving at dynamic percentages by dividing 127 by approximately 7 (from about 19% for ppp to 100% for fff), but the dynamics that result are unlikely to coincide with those of real performances. Sibelius's default percentages (from 34% for ppp to 100% for fff) are much closer to the real world.
Mozart only used three dynamic markings, p, f, and fp. The next generation added pp, mf, and ff. So we have very soft, soft, medium, loud, and very loud. A least one book on orchestration suggests that these five levels are all the basic levels that orchestral musicians recognize, since medium soft and medium loud are basically the same (and indeed mp doesn't show until much later than mf). fff and ppp don't show up until Berlioz.
In the real world fff is a dangerous marking because brass players often take this as license to push their instruments to levels that create a brassy, somewhat vulgar sound. (OK, if you want that.) And ppp suggests a barely perceptible level, yet all instruments have a limit to how softly they can play and still produce a sound, so the real-world dynamic range is narrower than the 19% to 100% scale would suggest.
The lowest level possible gets complicated. No oboe player can play C4 as softly as a clarinetist, and no woodwind player can play C4 as loudly as a trombone player. And there are some strange phenomena -- for example, a string section can play both louder and softer than a single player (there are sound acoustic and technical reasons for this).
After much debate in the 1970s, a notation conference held in Ghent, Belgium decided that the notation of dynamics in scores should be absolute, even though everyone knew that the first trumpet's ff was louder than the first bassoon's. The reasoning was that somebody (the conductor) needed to be able to look at the score and make sense of the markings. Thus the conductor could know that marking the trumpet one dynamic level lower than the horns indicated that the trumpet was truly supposed to be softer than the horns, and was not an attempt to make them come out the same. (With relative dynamics chaos develops. Some trumpet players can play much louder than others -- how would you compensate for that in marking a score? And how could the conductor know what you really wanted?)
Thus in notating a score, you need to mark it so that the score reveals the balances that you want to achieve. In rehearsal the conductor may need to tell certain players to play louder or softer in order to achieve this. But in using GPO playback, you may need to set MIDI playback levels that are different from the notated dynamic symbols. (Beware if too much of this is required because you may be asking the impossible of the performer. I once saw a Webern score that asked a soprano to sing a B6 pp, and some performer had written in the margin "Yeah, right.")
My main concern is that people might get used to an unrealistic dynamic range if they use the 19% to 100% scale. When I used this scale to playback the orchestral piece I mentioned via MIDI, the first comment my wife made was that the dynamic range was too large. She is not a musician, but she immediately recognized that the range of dynamics she was hearing was different from what she had heard in live performance.