Given we know very little about the problem and runtime constraints, the second approach has the potential to have better performance, as with case 1. the average duration of the requests in a batch is equal to the worst case for all requests, while in case 2. the average duration of all requests is, well, the average.