I dug deeper into the issue. I verified that the second HiveServerHandler does exit the method, just via an exception rather than via the return. This prevents the HiveServerHandler from being created. The test actually gets stuck making the first execute call on the second HiveClient, presumably because there is no handler on the other end of the connection to handle the request, so it gets stuck waiting for a response.
I admit I am equally confused as to why this started showing up so frequently recently. I had seen this problem a few times before, I had assumed it was caused by running the tests twice on the same machine. Now however, I've run into this problem every time I try to run tests, since yesterday.
As far as I can tell, there is no, and never was any logic to handle this problem. So Navis's diff seems worth committing to fix that issue, and it is at least a plus that it gets us around this issue, even if the root cause of why this issue suddenly became so prominent is still unsolved.
I would like to commit this patch, unless you have any objections.