This is pretty easy in cycles, by mixing a transparent node with an emission shader based on how close the camera is to facing the normals head on.
You can get the "facing" value with a Layer wieght node or by taking the dot product of the normal and the incoming rays from the camera:
These methods are pretty much equivalent, though you have a handy blending factor if you use the layer weight node.
Plug this into a mix node:
And you get this:
To get rid of the geometry behind the rest of the model, you can use the Transparent depth output of the lightpath node (note that this output is only available in the latest development versions of blender, but it will be in 2.71)
This value goes up based on the number of "transparent bounces" (a transparent bounce refers to the sampling ray passing through a transparent layer/surface of the mesh). Since we only want to see the top layer, we can use a Less than math node and a mix shader node to make every bounce past the first bounce (bounce 0) will use only the completely transparent shader (Transparent Depth is never negative, so this is basically testing if it's equal to 0).
If you want more control, you can use a ColorRamp node to fine tune the blending as much as you like:
Here is an example .blend for you to play with.
Note that I hid some of the input/output sockets on the nodes with CtrlH to conserve screen real-estate. So they might look different.