Digital implementations also often suffer from being too perfect.
Any motor-based oscillator can be accurately emulated by a digital oscillator, so the answer to the original question is "nothing." But yeah, neither is good enough to emulate a trumpet played by a human.
For synthetic sounds however, at this point I think it's mostly a question of workflow. Like yes you can take time to program all those imperfections into a digital sound. Or you can take time to make an analog device which has them inherently.
How do you know that every trumpet you've ever heard that sounded real wasn't synthesized?
They literally perform worse than digital oscillators, and that's desirable. Pure sounds are boring, sounds with character are what makes music interesting!