I suspect I might have unraveled what's going on between Pythia8 and the embedding.  Consider this post:

https://drupal.star.bnl.gov/STAR/blog/dmawxc/update-03242019-run-9-embedding-particle-level-pttrk-distributions-pi0-and-hadron-trigge

Now, below I compare the charged jet distributions for embedding, Pythia8, and data.  I've triple-checked everything to make sure all the jet-finding parameters were the same.

Which are the same distributions (as we would hope) that led us on this wild goose chase:

https://drupal.star.bnl.gov/STAR/blog/dmawxc/update-01072019-run-9-pp-detector-level-pythia-vs-embedding-vs-data-jet-spectra

But now this all adds up: the trends seen here (between the Pythia8 and embedding distributions) are very, very similar to what we see in the particle-level track distributions.  I would expect there to be differences between the charged hadron-triggered and pi0 pTtrk distributions because the charged hadron triggers include secondaries.  Moreover, I would anticipate this difference to propagate to the jet distributions; especially since we've been working with such small radii (R = 0.2).  That's what we see here.

The fact that the embedding distribution describes the data better than Pythia8 also makes sense.  Our pi0 selection isn't perfect: the pi0 sample is going to be contaminated by secondaries.  This may be a small effect, but -- to my knowledge -- we don't correct for secondaries in the pi0 sample.

Now, of course, we need to figure out what to do about this.