The reason I'm seeing that the DB first idea wouldn't be as fast as Gilles' way is that placing DB ends up being unnecessary.
- Orient edges after placing DB - About a 6 move average.
- Permute remaining 5 edges - 7.5 move average.
- Orientation step in normal LSE - About a 6 move average.
- Place UL+UR and finish M - A little less than an 8 move average.
Now what is left to consider? That you placed DB in the beginning - About a 3 move average. I also didn't consider the 3 AUFs for DB First, but I did consider them for normal LSE because it's really all one step. It's easy to average 13 moves for normal LSE in a speedsolve, that seems very difficult to do with DB First.
This is all just move count, not considering what is easiest to recognize and what type of steps are faster.