Optimizing parallel HTTP request batches