Well, lets see if I can answer all of these:
1. "Direct Y" is what I use, which tells it that the spring support is in the Y direction. This then pins the footing in the X and Z directions. "Yonly" literally only means Y direction only. So if you have any shear defined, the mat will fly off to infinite, so to speak. I use the command "Spring Tension" to tell it that the springs are tension only. This way the soils wont be modeled as capable of holding down the mat. With the tension only command, it will iterate until it converges. However, it might not ever converge, which will tell you your mat is not large enough.
2. I use Plate Mat. One of the tutorials I believe it is, describes the use of plate mat vs elastic mat.
3. Whether you include the walls for the stiffness is up to you. I would say it depends on the geometry. I have done both, though not on the same mat, so I can't compare the effect.
4. You can change the input units for that you enter your value of 75 pci, rather than converting to KSF/ft. This is what I do, as I have a feel for pci, and none for the other units.
5. Yes, I have modeled plates this thick. I think I have gone as high as 4 ft. Like I said, we were spectical at first, but after running some verifications, we felt satisfied with the results. I strongly encourage you to do the same. At the very least, it can give you an indication that you have modeled the mat and springs correctly. I ususally try to keep my elements about the same width as the depth. THough I do add more in tricky areas. I have used both 3 and 4 noded, again to get a feel for it. I usually use 3 noded, as that is what the autop mesher creates by default using the new parametric plate modeling function in 2004. This function is great as you define the corners of the mat, and any internal nodes you need for loads, walls etc. Then it auto meshes for you, including all the internal points you defined.
Hope this helps clarify a bit.