… monitoring plugin objects that are exchanged with other cluster nodes in task execution

Creating the JAXB context might be somewhat resource intensive. As we're lazily initializing things, we'd better guard against multiple threads trying to do the same.

Pet peeve. I've battled to many bugs that would've been very easy to fix when some kind of unexpected exception wasn't ignored.

Much of the code that depends on this declares IOExceptions to happen when serialization fails. We might as well make use of that.