Again, you may still be wrong. Many systems have different strategies for large blobs vs. small blobs (I know it's many because I've written several myself) including chunking strategies, separate memory pools etc.
I'm taking this stand because I've had the exact issue we are arguing about: my large file handler had a leak in the memory pool for large blobs. Basically it was shared memory space for multiple processes so large objects could be passed out of band instead of using IO.
Pretty aggressive stance when you may, in fact, be wrong.