Interesting read, but I am doubtful about some aspects. For example "SQL Blobs are the wrong choice for binary data" - wouldn't that depend entirely on how the SQL Blobs are implemented? Sure, maybe apache can serve faster from the file system than from an SQL Blob (if you read the SQL Blob in the middle layer and stream it). But that is just one possible implementation. In theory, the web server could be part of the database and stream the SQL Blobs directly. SQL Blobs could be implemented as files in the file system. And so on...