What I think would actually solve this is a clock that doesn't schedule in real time, but as fast as possible, a server that doesn't render sound, but just records its OSC messages (in this case, it would also have to collect the 3 SynthDefs that are created as well) and then render them.
That sounds like just the ticket.
This might be doable, but I would also argue that the code below is not really efficient SC3 code and most people do not code real time work in this style. There are few if any examples that don't show SynthDefs and Synth objects.
I can accept that, but even expressed with SynthDefs and Synths, the RT version of my example would still look much different from the NRT version. Further, one can expect many users to not use the most efficient means, as long as the result actually works on their computer. (I can attest that the example I posted plays on my 2 GHz MacBook without glitches.) A general NRT rendering scheme will work for any legal SC audio-generating code.
The change you mention is a SuperCollider 4 project IMO.
That would be most excellent, if enough developers agree with me that it would be a good idea.
Cheers,
Eric