YES!!! I just mapped my small PK/PD library which is formulated using ragged arrays to use rectangular data structures. That means I should rather quickly be able to run real project problems with this approach. The key was to figure out a conversion from ragged structures to a (padded) rectangular data storage.
@Bob_Carpenter: For me exact repro is a really important thing and we should by all means make it at least possible that exact repro can be warranted. Moreover, the MPI stuff works by chunking the N jobs onto the W workers. Often N >> W as we do not have a gazillion of workers. Since W won’t be too large, the gain due to asynchrony may not be huge.
In pharma it would not be acceptable to have not exactly reproducible results. So if we can make this optional (with a performance hit then), that would be the best approach. I would expect that other fields are similar in this regard.