The "unification" was something that was emphasized when GUT's were the cutting edge of high energy physics. The idea then was to make the interactions come from a nice symmetrical group. The point of physics is not a unification in this sense--- making everything be one thing in this way. Rather, the point of physics is unification in the sense of a consistent mathematical description that describes everything, no matter how many parts you have. To unify gravity with other interactions means that they are consistently described together, not that they are fundamentally all due to one kind of particle in a low-energy theory, or that they come from the same gauge group (this is the misguided interpretation of unification in some recent work, like the exceptionally simple theory of everything).
The GUT unifications, which unify the non-gravitational forces into one big group, are important clues to reproducing known physics, because they tell you what kind of symmetrical structures you can expect to break to get observed physics. This is turning the low-energy theory into a useful clue about how to produce it from a high-energy description with a lot of symmetry.
With the advent of string theory, we have such a high energy description. The description in string theory is always unified in the second sense, it always comes from a consistent theory at high energy. The GUT unification then tells you what kind of string vacua you should be exploring to get observed physics. It is the case so far that the usual string models which reproduce something close to the standard model do so by breaking a GUT slightly, so that the low energy theory is unified at intermediate energies. This is true even if you make the standard model on branes, because you have extra symmetry when the branes are on top of each other.
The dualities in string theory are not unifications exactly, although I guess you can think of them as unifying gauge descriptions with gravity descriptions in some sense. they are precise formulations of string theory in different domains, using the fundamental degrees of freedom of different objects in the theory. The idea is not that you get everything from a single group or something, rather it is to describe the dynamics of string theory precisely using known non-gravitational models. The original formulation of string theory was also a kind of duality, a duality between 10 dimensional space-time physics and 2 dimensional world-sheet physics.
The description in string theory is holistically consistent--- it comes from a structure that can't be tweaked without spoiling everything. This makes the string theories fully unified from the start, since every part is required for the consistency of the rest. The strings also have the property of "nuclear democracy" (which is really gravitational democracy), which means that you can consider everything as made of strings, or also consider it as made of of certain D-branes, and you get the same description no matter what thing you choose to call fundamental, although with different limits becoming easy to calculate.
But string theory isn't saying everything is made of one specific thing, rather it is saying that any one of the things can be used to reconstruct the rest of the things. You can see more here: Why do we have so many dualities in string theory? .