I think it's better to introduce a language dependency instead of extending an existing language with new features if those new features won't be a natural fit for your existing language.

For example, GADTs are natural in strongly+statically-typed languages (e.g., Haskell) because they're typically used to write code which you want the compiler to check for correctness (as much as possible); if your existing language is a dynamically-typed language (in this case, Python) the interpreter may still be able to enforce type-safety at run-time, but you'll lose the more practical benefit of knowing at compile-time that the code won't break. Adding a feature like this to Python would add additional complexity to the language while also taking a significant amount of time before it was supported well-enough across the entire range of Python environments to really catch on; and for all that hassle -- which you're going through just to stay within Python, remember -- you're receiving minimal gain, and it's likely to cause more headaches in the long term than it solves.

FWIW, I'm actually quite a big fan of Python, even though I don't use it much; in particular, the Python open-source community is excellent. I just think that languages should focus on solving whatever problem they solve -- it's easy to get carried away with adding "more power" and end up ruining something which already worked well for a particular problem.