Describe that service implementations can process adjacent GET requests in parallel as long as the responses are represented in the order the requests appear in the batch request. Only change sets and stand-alone data modification requests have to wait for preceding request to finish.

This might require invisible contracts between the server-side service infrastructure and the actual data provider code as to whether the supposedly non-side-effecting GET requests really can be parallelized, but this is nothing that shows up in the protocol or $metadata.