This post just brushes off synchronization and caching. For a trivial example like this yes, it's fine, but in practice this is much more complicated. This really only covers the "embarrassingly parallel" algorithms, which are the exception, not the rule. If you were to generalize this, you would end up with something like https://github.com/NVIDIA/cccl/tree/main/thrust, which you can click around for a bit and see it's far from trivial

Also, cache aligned densely packed data is what CPUs were created for, very often, maybe always, you'll get much better performance by properly organizing your data in such way the CPU can use its caches to their fullest than just throwing threads at it (of course, you can do both)