mongodb::Cursor is in fact a RawBatchCursor (source) wrapped in a stream encapsulating a VecDeque<T> buffer (source). The single-document-based iteration API of mongodb::Cursor forces intermediate allocations and deep-copies of every individual BSON document within the full batch into this buffer. The CursorT implementation should be refactored to bypass mongodb::Cursor in favor of directly accessing the raw batch document instead, avoiding the overhead of intermediate copies per document in the batch.