The problem is that the constraints of wanting your data and CPU on the same node, needing data synced to more than one node, and avoiding being too much of a p

The problem is that the constraints of wanting your data and CPU on the same node, needing data synced to more than one node, and avoiding being too much of a performance bottleneck, all combine to make it very hard to do better than something which has privileged knowledge of how the database works under the hood. Performant consistency is hard.