I just uploaded a video showing the differences between typical method comparison regressions when outliers with leverage are massively present in the data.

Here: robustness video

Bayesian Deming methods turn out to be particularly effective already at df=6 or df=5 for the T-distributed error term setting.

Nonparametric methods are not immune to coefficient shifting due to outliers and are less robust than expected.