It seems like it might actually be a bug if the synth was that non deterministic that it generated output that was so drastically different to be audibly different. Maybe the developer intended it to be this way - I'm not sure.
AAS (Tassman) would be surprised to hear that! One of the things that makes Tassman so lifelike, as well as enjoyable to play, is that is responds like a real-world instrument; in part, I believe (without knowing) this is because the programming includes some of the same "chaos" that exists in the real world. You can blow into a flute 5 times "the same way" and the air will move through the tube differently each time.
The example of music notation as opposed to recording an instrument applies: Sure, every violin playing a notated part will sound more or less the same, but every performance will be a bit different. SONAR users wouldn't settle for the answer, "Well, you can save the staff notation, it will always sound the same when the fiddler plays it back."
A soft synth these days can be so much more than just a device that captures and replays MIDI data!
Two more examples: (1) Reaktor includes a bunch of noise/sound sculpting synths, emphatically based around random elements. A player can improvise with these synths, and midi can capture the way sliders and knobs move, but since the instruments are BASED ON randomized elements, replaying the midi file won't recreate the same performance.
(2) My keyboard includes channel AT. I generally have this turned off in global options as a recorded parameter, per CW's wise recommendation, since it generates so much MIDI data. But if I'm improvising on a synth that happens to respond to AT, but forget to reset my global options, I end up not capturing that aspect of my performance. (This has happened to me!) Routinely recording audio as well as MIDI would safeguard against this.
CW provides several different kinds of dither. I bet the majority of SONAR users can't hear any difference. Is it a bug if different dithering algorithms give audibly different results? The point is, many of us do sweat the smallest details -- I bet that's why CW gives us dithering options, since some users DO care about things most people can't even distinguish.
Oh well, I've said my piece. Merry Christmas to All!