I'll briefly answer the above questions, but we could we close this issue?

If there are specific issues regarding PEP 626, please make a new issue. Feel free to +nosy me on those issues.

I intend to review all the recent changes to the compiler in the new year,
in case there are any obvious inefficiencies.

Ned,
The bytecode for 3.10a should be more efficient than 3.9.
IMO, it is not worth spending too much time on making the bytecode look pretty, and that if we care about performance it would be better to spend our time elsewhere.

Serhiy,
The bytecode for the test in the while statement is duplicated; one in the loop header, one in the body. Try it under pdb, you'll see that it breaks at two different places in the bytecode. 
See https://github.com/python/cpython/pull/23743

Benjamin,
Eventually, yes we will want to implement something like the JVMTI for the JVM. However, for that to be useful we still need precise line number tracing.