Currently tests are all run in serial, which results in suboptimal test running performance (in the case of umoci we've started to hit against the 50 minute limit that Travis has because we run ~200 tests that all take a fairly long amount of time -- but are easily parallelisable).

There was an issue about this in sstephenson/bats#143, but I think the easiest solution would be to use GNU parallel if it's available (which knows how to collate data between different programs and can also do a bunch of other clever things when running). In theory the fix should be quite trivial (a minor rewrite of bats-exec-suite to use GNU parallel for running bats-exec-test).