Your setup and problem seems similar to ours, but at a slightly larger scale. For us, the problem was not so much ember-cli as it was the Mac itself.
Once we got past the natural ‘ulimit’ problems that prevented the build from actually succeeding (solved by upgrading OSX or by manually raising the ulimit) we found that the file read access was being shared between node (no watchman, here) and the Mac indexing service. As node_modules, bower_components, tmp and dist were rebuilt, Mac was doing its level best to re-index for us - entirely unnecessary. The issue was exacerbated for me because I use Webstorm as my IDE, and it maintains its own file index cache.
I don’t know how the results scale, but it dropped my build from ~13 seconds to ~10. I didn’t dig any farther because I’m not far removed from some serious Java projects, where we were looking at 5-7 minutes per build/deploy…13 seconds seemed like a miracle!
I hope this helps a little!