It's pretty common for products like this to first make your implementation work correctly on a single node then scale it out to correctness on several nodes. Sounds from the blog post like they're midway through that process. It's still in beta so it all seems pretty reasonable to me.