True, it is a more natural progression to parallelize on the backend as isolating different requests has multiple benefits besides scalability; e.g. security and maintainability. Also, we have more control over the backend environment so it's easier to add external components (e.g data store like Redis) if needed to help keep track of state; there's no need to store everything in process memory... Not to mention that most backends need/use a database already and so it can be used to keep state with relative ease and efficiency without even adding any new service/component.