How does it depend on the depth test, and why does it seem "off" to make
something that is originally a shader property in D3D a shader property in
OpenGL as well ?
Not using shaders isn't allowed (at least not in gallium), so you have to use a
pass-through shader anyway, so this limits state inter-dependency.

The qualifier "layout(windowspace)/layout(pretransformed) out vec4 gl_Position"
is supposed to disable clipping and the viewport transformation (so glViewport
settings will be ignored).

The only open question is how depth values should be treated, which can be
either clipped or clamped, and there is actually a D3DCAPS9 bit to tell the
user what will happen.
On r600, you cannot disable xy clipping without disabling z clipping, so it
will have to advertise clamping. On NV cards you have more fine-grained
control, and I'd actually prefer depth clipping there unless GL_DEPTH_CLAMP is
enabled as well.