To index millions of PDFs/images quickly:
Partition the blob data into virtual folders (prefixes) and create multiple indexers , one per folder, so they can run in parallel .
Increase search units (SUs) to raise indexing throughput.
Schedule the indexers with the same cadence so they run concurrently.
Creating multiple services (A, C) is unnecessary and costly. Scheduling sequential execution (B) defeats the goal of "as quickly as possible." References Azure Cognitive Search indexer performance guidance: parallelizing blob indexing using multiple indexers over virtual folders/prefixes and scaling SUs for higher throughput.