For M2, DB as buffer makes a lot more sense, because for setup moves instead of doing B moves you are doing F moves, which are easier to finger trick. I really don't see why you would use DF/FD, except that that is how the method was developed. But if you are new to learning then it makes a lot of sense to learn it that way. Are there any cons to DB/BD as buffer?
1) Your buffer pieces is hard to see, which means you will need to spend some time peeking at the entire DB piece to determine how you're starting your memo. DF is much more visible.
2) DB is a worse buffer for 3-style, which is something that you will want to switch to in the future. You want most of your algorithms to involve "nicer" pieces, i.e. ones on the U/F/R layers. This makes DF better than DB as a 3-style buffer.
(However, UF is a better buffer than both of them, and UR or UB are good choices too. If you want to switch to 3-style, you might find you want to change buffer anyway, so this second point would no longer be a disadvantage. In fact, the M2 algs for DB buffer work well as comms for 3-style UF buffer, so it might actually be an advantage of using DB for M2.)
3) Fewer resources are available, though everything is quite intuitive anyway, so this shouldn't be much of a problem. It certainly won't be a problem in the long term.