There's a tremendous difference between standing in an audio booth recording lines and actually being given the space to perform. Letting Cumberbatch get into being Smaug, letting him stalk around, leap on piles of "gold" or play with a trinket, will change the way he delivers the lines. The rhythm and intensity and inflections will all be different. And while an actor of his caliber could no doubt fake it just fine standing in a booth, the animators at the other end would have no idea what was in his mind when he paused here, huffed there.
So you put him in a space and let him do it, and at that point there's really no reason not to gather mocap data too, just to have it. It would have to be retargeted, yes, and possibly much of it discarded as inapplicable, but certain nuances you might gather -- the fractional tilt of his head, say, or a small hesitant motion of the arm -- could make all the difference in translating Cumberbatch's specific performance to the final animated character.