I also wrote something (before confd but I've yet to see anything that replicates it) for treating etcd like a filesystem and doing edge-triggered updates. You form directory structures, and those long keys get mapped to files on disk with their value as the file contents.
What really bugs me about these solutions -- confd, consul templates -- is that managing templated configuration files isn't as quick, but has been trivially possible with normal configuration management tools since their inception. When I think of managing HAProxy configurations, for example, I don't see Chef as a bottleneck. I can parallel SSH and run Chef on those hosts in 20 - 30 seconds. I don't care about the configuration being pristine, either.. because that's what backend health checks are for. That extends to a lot of their examples, but it's one of the ones that stuck out to me.
The things that I think are hard, or harder, are small bits of data, like database credentials, etc.... where people absolutely don't want to be making an HTTP call for every single thing, or don't want to define a template for hundreds of individual files that might be needed. That's a real bottleneck - you need something on top of these solutions to manage that. So for every file you care about being updated fast... you still need to manage a corresponding file.
I don't know necessarily how to solve that -- I treat K/V pairs as file name/contents pairs in my tool and don't think twice -- but it definitely feels a bit wrong.