People have got very realistic sounds out of some synthesis methods - additive and physical modelling are two methods that are flexible enough to produce very realistic sounds.
However, both techniques are difficult. If you want to create a realistic sound from additive synthesis, you have to control in detail the volume and pitch of dozens (maybe over a hundred) of individual sinusoidal partials as the note evolves in a way that reflects the instrument being emulated. That's just for one note - and then you might find that another note, or a note played with a different technique, has a different behaviour. Then a performance made up of many notes has other aspects to consider... it becomes quite an undertaking.
When it comes to physical modelling - If you have a good mathematical model, then hopefully that takes care of the 'instrument' behaving in the right way - but you then have to 'play' it with all the same skill, nuance of 'touch' and all the inflections that would make a good performance on a real instrument. I have real guitars and I dabble in simple physical modelling synthesis - I find that creating a performance on a real guitar is a lot easier.
Some physical modelling techniques are a bit like what you describe in terms of being a network of many small elements - have a look into 2-d waveguides, for example, or the NESS project.
See also - Can a physics simulator be used to generate sounds for music?
I'm not sure how much of electronic music is completely synthesized sounds vs. samples
It's a mixture - in general though, the fully synthesized sounds you will hear are still mostly identifiably 'synthesizer'-type sounds rather than convincing simulations of (say) acoustic instruments. Although there is a reasonable body of human knowledge about how to make realistic additive and physically-modelled sounds, the issue until recently has been computing power - but affordable computers are now becoming powerful enough to run quite complex physical models and additive engines in real time.