Moog's iPad apps and the new UAD Minimoog both sound pretty good, so there's no practical reason why DSP synthesis should sound bad.
The only thing preventing that being used here is the complexity of the models and the fact they tend to be proprietary. Supercollider, Max, Csound, PD, and ChucK all use the simplest possible public domain models, and even the best of the simple models aren't good enough to generate sounds with the production values you'll hear in a commercial recording.
Generally the computer music community doesn't pay attention to production values at all. They're happy with a workable basic implementation of $effect or $synthesis_element, and they lose interest in the rest.
That's partly because DSP with high production values takes a long time and work. There's a story that gets told about the TC Electronic 6000 system, which has a gold standard reverb you'll hear in most movies [1] - which is that it was the work of a team of more than ten people working for years.
It's not hard to hack together a basic reverb algo from public sources. But it is not going to sound that good.
[1] You know the ambient piano sound some movies have in their scores? That. If you want to buy that reverb the hardware costs nearly $17,000.
