I can see your predicament but the approach that GLSL takes towards scanlines does not suffer from the same issue. I believe this is also what Roger is referring to.
I understand that the link above can be argued to be less accurate than the simulation that HLSL provides. However, the scanline and aperture aspects of the GLSL seem to scale correctly.
Would it not be feasible to adopt a similar approach for HLSL whilst keeping the remaining, more accurate aspects of CRT simulation intact?