According to our data, our ISO builds are not slower recently than they were
before we switched to Vagrant, go figure. But we've lost the data from the
months immediately before the switch, so it's hard to tell whether I'm comparing
apples and oranges or what. Also, I guess we probably optimized other stuff
which compensates the nested virtualization overhead. Now it's too late to
measure the difference between the old & new setups when doing the exact same
job, so let's just give up.