I've pushed a few minor changes to the README text. BTW, might want to
consider making this a markdown file.

According to the README, when tests are running in parallel, output is jumbled
together in systests.output. This seems like a bad idea; let's capture the
output of each test separately, then concatenate them as the tests complete.
I'd put the output for every test into <test-name>/test-output, and then have
stop.sh "cat <test-name>/test-output >> systests.output". (There's still a
slight chance two tests would be writing into systests.output at the same
time, but it would be much reduced.)  The top-level clean.sh can remove
test-output; that way it won't be necessary for each per-test clean.sh file
to do so explicitly.

A minor feature request: I've often wished there were a way to run a subset of
the tests with a single command, like "sh run.sh dnssec autosign inline",
but never got around to implementing it.  While we've got the hood open, can
we do that?  If we support "make -j3 test-dnssec test-autosign test-inline",
we can even take advantage of parallelization.

There's nothing in the README about the SEQUENTIALDIRS variable in conf.sh.in.
IMHO that should be included in case someone needs to write a test that
can't be parallelized for some reason.