Mike taking a break aside, I thought I would help clarify the geometry situation a little by using actual 3D models. All of this applies equally whether you're using the top-down shading Meph has suggested or the directional shading that Mike has been using so far.
There are five different possibly configurations I've come up with. I will be using the same exact same landscape for each one to make the differences more obvious. The tiles are checkered so you can see which faces belong to each tile.
1. I'm starting with this one because it
does not work.This is what that would look like in 3D, and obviously it's nonsense.
2. The geometry on this one actually works. It has smooth slopes along the horizontal, but jagged slopes along the vertical. The disadvantage is that it makes half of the corner ramp tile solid wall, which would look weird with a sprite on top of it. I've converted Death Dragon's mock-up into a gif that shows what it would have to look like one z-level down.
Here is the 3D:
3. This the option that most of Mike and Meph's demos have used so far. It is the opposite of option 2 in the sense that it has smooth vertical slopes but jagged horizontal slopes.
Mike's v3:
Meph's v3:
3D model of v3:
4. This is a hybrid. It uses the convex corners from v2 and the concave corners from v3.
5. This is the opposite hybrid, using the convex corners from v3 and the concave corners from v2.
Bonus Version 6:
If you don't care about keeping things planar, you could go for a series of saddle shapes. The problem though is I think it's hard for the human mind to get a good grip on this just based on shadows and light, and it wouldn't work very well in 2D because of this.
In conclusion, I think version 3 has to be the way to go. Version 2 works equally well in 3D, but it doesn't work in 2D tiles because those concave corners put half of the sprite 'in the wall'