I don't think there was any "Roman average". We have "gladiator" or "soldier" samples from the Roman era in Britain, and they're pretty similar to modern day Brits and Dutch, Belgian, and northern French people. The one "outlier" is clearly from either the Levant or Arabia.
In some areas you do have Roman colonies of veterans, so you would get some admixture, but after a hundred years the signature would start to wash out because of intermarriage with the much larger surrounding population. Anyway, there don't seem to be very many in Pannonia, at least not in the second century, so that part of my speculation upthread is probably wrong.
I think it's far more likely that these "Italian" plotting people are the "native" people of the area. This was densely populated EEF territory. There wasn't the large effect from the Indo-European migrations that we see further north in Europe. Remember the results in Mathiesen et al, where even in the Iron Age the "steppe" component in most of the samples was about 10%, and some of the samples also looked quite "Tuscan" like. Let's not forget that the GAC had many of the hall marks of an Indo-European culture, but they were just MN European farmers genetically. Language change apparently took more, but 10-20% seems to have been enough.
The closest ancient population to modern Italians is often the Hungarian Bronze Age, which also isn't very "steppe". Actually, a lot of this is starting to fall into place for me.
As for the "CEU" type samples, I'm not sure why the authors are surprised. The area had experienced its own "Celtic" migrations, similar to what had happened in Northern Italy and all the way down to Rome. I would think there might be "Celtic" settlements, as there was a "Lombard" one.
https://en.wikipedia.org/wiki/Pannonia