Jeff, this is a LONG Shot..... but exactly 6 years ago (to the day!) I recorded the same thing with Borons. Turned out that the Device Under Test would start communicating with the Cloud 4 seconds after a different Boron on my "test" Particle account sent a publish.
There were no Particle.variables, Subscribes, Functions, etc.
Here's my summary a couple days later. Physical proximity wasn't related, as mentioned later in the thread.
My suspicion was something on the Cloud Side changed on 12/18/2018 which happened to be during my test, so I had specific documentation of the results pre and post.
It might be worth a few minutes to perform your same test on a temp Particle Account with only 1 Boron? Like I said, it's a long shot 