Yep, unfortunately this is not something that could be fixed easily, as you need to make sure mapping the multiple observed RVs correctly across models. As a work around I suggest you do sample_ppc for each model, and then resample the ppc samples according to the weights.
Also be aware that depending on what kind of weight you are using, some of them might not even make sense for multiple observed RV or multivariate observed. I think loo based weight for example is mostly only for univariate observed.