The Linux times stayed at ~2.5min. Any idea why? What I just saw on this run was that it hung on INFO: Computing changes... a long time (although maybe I miss-remember this up as the log suggest differently).
It might be related to recent travis macOS problems, but then I would have thought it would affect wait time but not run time.
I advocate treating Travis for OS X as a scarce resource: if you have a good reason to use it (eg a package which is not pure Julia, has some tricky cross-platform stuff, etc) then do, otherwise disable testing for OS X until the situation is resolved.
This would result in faster test times for those who really need it, and obviously faster test times who skip these tests. The downside is not catching some subtle bugs which could crop up nevertheless. I recognize that more testing is always good, and suggest this as a practical compromise given the constraints.