Adjust to Touch can move surfaces in unexpected ways. Look in the Solution Information folder for Solver Output, which is the listing of the Solve.out file. There should be a statement on how Adjust to Touch moved the contact surface in X,Y and Z. Compare that with the offset distance you are using. You can move the faces both ways in CAD and see the effect.
How are the two sides of this joint supported and loaded? Does that support/load allow the joint to "find" the natural, lowest energy seating orientation? In other words, if the top side is fixed, is the bottom side free in X, Y, Z and Rx, Ry and Rz to adjust so that there is only an axial load on the bottom side? If there are too many constraints on the bottom side, then unrealistic torques will be supported while the joint makes contact at high spots, rather than at the lowest point.
What is the experimental data measuring? How is it measured? What is being measured in the model? Is that the same measurement?