I think mostly I'd use microframeworks like this in the future to do small projects and early prototyping.. or if I thought the project had special needs that was going to make using something like Django a particularly challenging match.
You've got to decide for yourself where your 'reinventing the wheel' cutoff point is. For some people it's flipping binary bits, for others, anything lower level than configuring Drupal modules is just a waste of time.
At a high rate, bottle just dies. So if you have a situation where you are making many requests to it, be careful; your mileage may vary.
So, why the asterisk next to "no issues"? I was able to wedge our app once by doing frequent, unbuffered, blocking socket.reads for one byte at a time in our RPC library... the GIL is a fearsome thing and it didn't play well with this strategy under highly concurrent request loads. I don't see that as a bottle problem, though.
If you're anticipating high loads, though, I wholly recommend looking at Brubeck (http://brubeck.io/) or Tornado (http://www.tornadoweb.org/)
I would have been happy to use Bottle if i) it had a feature that lets me specify FDs to watch and associated callbacks, like what I could do if I directly controlled the server select loop and ii) Python had built-in support for reader-writer locks.
Twisted had async support but it's documentation could be more comprehensive. I understood the programming model, classes and methods well enough to use it for my project only after tinkering with code, not from just reading documentation or looking at examples.