the single line itself
It main purpose was that it's executed within the JVM on hadoop-qa. That's how we found that our ulimit were not taking into account on hadoop-qa. I now use it to see that the test has actually started & finished in the logs. There's no impact on test coverage if we don't have it. I would be lazy here I think: don't spend time to remove it when it works, don't spend time to make it work when it doesn't. We fix/remove it when (if) we need it.
hbase-common all run in the one JVM
It would work short term imho. If we could, it would be better not to have this as a prerequiste, we can imagine having a lot of stuff there, with maybe long running tests or tests playing with the environment (typical stuff: using defaultEdgeEnvironment to manipulate the time needs to be done in isolation as it's a static). In the same area, if we share code around thread management, with extreme/error cases (too many threads & so on), it's better to do it on another jvm. I guess there are other example
Yes, really the main advantage is that it runs on hadoop qa as well.