I was in a similar situation once. Decided to move our BLOBs to a flatfile system and use the database as an index to our new file's archive. Your blobs are actually not that big, so filestream won't make much improvement.

What's your read/write ratio? how often are your BLOBs being access and created? what's the ratio of BLOB queries vs regular queries?

Looks like you would benefit from a clustering solution in active/active state where you would query your BLOBs on the secondary node to reduce the bottlenecks in your application (if any).