We've been driving all of our Ember tests with end-to-end Capybara tests running from our Rails server for some time. There are many upsides to this approach, but Capybara tests are generally slow and brittle, so we'd likely to start investing in client-side only Ember tests that don't actually hit the server at all.
Given that I don't have a lot of experience with Ember tests, I wanted to keep a journal of first impressions and snags I encountered along the way, so that perhaps this information will be useful for documentation maintainers or whomever.