I'd recommend a filestream over this approach. Filestream is natively supported, and offers asynchronize I/O, which means inserting your blobs will be faster, as will retrieval. Filestream is ideal for storing files of 1mb or larger though, and anything less isn't worth the overhead. I use it to great to store certain dataset raw files, and the performance gain of using filestream was about ~4x the performance of trying to do it without it.