Really depends on how much searching and organizing you’ll need.

I personally prefer the idea of using a file-key in whatever DB I’m using that way I can experiment with different storage methods while the actual DB lookup keeps my static value. You’ll have to code something to load the actual media from the key, but you won’t have to worry about the two being tied together permanently.

Example: (I personally use mongo for my DB so everything is a JSON)

{ "_id":201,
  "title":"kitten fights lemon",
  "author":"CowFu",
  "status":"Published",
  "tags":["kitten","lemon","fight"],
  "thumb":"p1000321",
  "media":{"type":"mov","key":"m874389"}
}

That way I can just lookup the media.type and media.Key against whatever storage I’m using.

I’m currently loading from Ext4 on debian, but am going to try BtrFS next to see if there is any difference.