The problem is that adding TCO to a mature and widely-used language introduces a massive compatibility problem!

  If Python supported TCO, more recursive Python code would probably be written

And not be able to run on Python implementations without TCO!

Code that takes advantage of TCO is utterly dependent on it. It is absolutely not an 'implementation detail', or really an 'optimization' -- its presence is a major part of the semantics of any language implementation that does it.