I wouldn't say Nim is "radical", but it is interesting in that it combines ideas in a way that I've not seen in other languages. Whether or not it really takes off, it's a good thing that it's around - if nothing else, it's a slightly easier way for devs who work primarily in higher-level languages like Python and Ruby to encounter concepts they might not otherwise encounter, like macros and a complex type system.
Few projects: nbuild[1], ntangle[2].
I wouldn't say that these are exemplary projects.. but they just show that Nim can replace bash/Python for command line utilities.
[1]: https://github.com/kaushalmodi/nbuild/ [2]: https://github.com/kaushalmodi/ntangle/
If performance isn't a big deal, the other thing I've noticed is that compilation guarantees a much higher quality of code compared to interpretation. At least compared to Python, I have been hit by bad code which didn't get caught because there was no test for it. I've not faced that with Nim yet.
No doubt you can still have issues but it's one less thing to worry about.
Probably a bigger factor is that Nim doesn't have a well-known organisation such as Google/Mozilla behind it.
I liked playing with Nim in spare time. But, I don't think I'll use it for anything I care about to maintain for a long time.
If you do want the new cooler features, then ya, you may need to port your code but that's the case with any compiler update - you need to run a full regression at the very least.
If that's not what the developers are trying to communicate with a pre-1.0 version I do wonder what they're gating 1.0 on. What needs finalization? Documentation?