Yes. The models position is determined by the pelvis body of the ragdoll. If you don't have any Kinect data yet, then the pelvis body is probably positioned at position 0 = on the floor. You have to position the pelvis body manually until Kinect data arrives. 

Visualize the skeleton of the model and compare it to the Kinect skeleton (which is already visualized). If the model skeleton is different from the Kinect skeleton, then you might have to apply a scale to either the Kinect skeleton or your model's skeleton. Maybe it helps to attach the "Marionette joints" to other parts of the skeleton. Or maybe you need to tweak the joint settings (e.g. MaxForce).

If your skeleton is very, very different from the Kinect skeleton, then it might be better to map the Kinect skeleton to the model skeleton using the SkeletonMapper class (see skeleton mapping sample in the CharacterAnimationSample solution).

Actually, I did not understand this line of code ;-). The pose of a body is given in world space, so something like this should work:

newPose = new Pose(oldPose.Position + new Vector3F(0, 1, 0), oldPose.Orientation)

Sorry, I don't understand this question.

You need to create a custom ragdoll for your model with custom shapes, joints and limits. (see this article for example: http://www.digitalrune.com/Support/Bl...)

In OnSkeletonFrameReady, there is a hardcoded y offset. Maybe you need to tune this value.

If the skeleton of your model is exactly the same as the Dude's skeleton, then you can reuse the ragdoll. Otherwise, it depends. I cannot answer this question for you.  

A difference in the skeleton movement or shaking ragdolls can have many reasons. Have you worked through our physics and ragdoll samples and the mentioned article? (I think you need tackle this problem step by step and practice the basics first.)

Do the bones in your model have the same orientations as the bones in the dude model? To find out you can compare the dude model with your model in your 3d modelling tool. Or you can visualize the bones with SkeletonHelper.DrawBones() in the XNA sample. SkeletonHelper.DrawBones() draws three lines (red, green and blue) to visualize the coordinate system of each bone. If you reuse the dude ragdoll, the bones in the dude model and in your model must use the same orientations.

To return to the original position. I am sure there are several ways to achieve this. One idea: When the model is in its default pose, remember the positions of the model bones which are controlled by the marionette joints. Later when the player was lost by Kinect, manually set the bones of the _kinectSkeletonPose which control the marionette joints to the default positions that you have stored before.