probably not. The moment someone has a module level import of from typing import SomeAlias the __getattr__ will be invoked. For this to have any benefit, you’d need import typing as t and then use of t.SomeAlias only in places where evaluation of the attribute access is deferred, such as type alias statements or annotations (3.13 and earlier with a future import, 3.14 forward with 649)
This could still be done, but it would require adoption across “foundational libraries” to make it likely that users dont end up eagerly importing anyhow. It also rules out use of typing decorators.
You can see this post for details on the current limitations of runtime deferral of typing.
Something that would improve performance here is to remove all the metaprogramming tricks going on in that file and expand it in source code instead. This reduces the amount of things that need executing at import time, but that’s not free either, it has costs on development in some form or other, and would probably make a codegen script appealing (even if it’s run manually and results checked in).