but I wonder how much of that is that they just had more variable cement, and the stuff that wasn't incredibly great turned to dust long ago? Certainly, our modern processes are more repeatable and produce a narrower range of results. What if the existent buildings are outliers of the roman process?
It's unlikely. They used a completely different chemical formula than we do. If you have two processes X and Y, and you observe consistently that Y >> X, and you know that Y and X are produced by different processes so you don't have a strong prior on the mean of Y being similar to the mean of X, then you should conclude it's very likely that the mean of Y is greater than the mean of X. Even if you know for certain that you only have the best samples of Y.
But more importantly, we don't really care if the romans also built a bunch of bad buildings using counterfeit cement with a botched formula, that failed to last until today. What matters is that we don't have the technology today that can match their best work. That's what makes it worthy of study.