I hadn't noticed this when I was working on my tessellated terrain, but looking back at some screenshots I took in wireframe mode, that tessellation scheme does show up.
At the time, I was working about half the time on a DX10-grade card that couldn't do hardware tessellation, so I wound up spending a fair amount of time doing an adaptive cpu-side LOD system. It's more work, particularly stitching up the edges, and the performance wasn't as good, but if controlling the tessellation is essential, that way you can subdivide however you choose.
Out of curiosity, why would the alternate tessellation be better for your heightmap? I don't really know anything about Havok
Nice work with the hieghtmap LODs, back in my dx9 days i did something like that. If i don'f find some hardware based solution, I'm going to use a similar method. I have all the heightmap data for physics near the camera, i upload them for the for the gpu, maybe only for the close chunks, and use the tesselation for distant parts.
Havok has a very simple hieghtfield shape I'm using, the 'hkpTriSampledHeightFieldCollection'. It only needs the heightmap dimensions,cellsize, and the triangle winding. So all the cell must be divided into triangles from topright-bottomleft or topleft-bottomright. It is quite fast to create, fully dynamic because it uses the heightmap data the user has, no need for uploading or preprocessing. My game is procedurally generates terrain, so i need all the speed.