Looking forward to implement sparse voxel octree code with iterative mesh refinement with Chapel. I forgot about it, because it was unusable for GPUs back when I first had contact.
And now it seems good enough for non-GPU code and maybe even GPU code in comparison to C++.
I expect a few 10s of TB ram to be required for a specific dataset, which practically rules out any shared-memory approach. Clusters should be better above a few hundred GB anyway, maybe less depending on network bandwidth vs. RAM needs.
Even if not used in production, Self is one of the influential ones on the list. Self did JIT before Java, did prototype object model before JS, and did visual morph objects that I'd guess influenced Squeak and others.