A "true" heartbeat would provide several answers to problems we're currently facing:

• it's not known where the "deadlocked" JVM actually is (stack traces),
• it's not known whether it's a resource (no cpu usage) or active (spinloop) deadlock, sending back the stack traces a few time could answer this (as well as maybe generic CPU load),
• it's not known whether the forked process is still alive at all (should heartbeat messages stop arriving, we can try to terminate the forked process and end with a build error).