For the CD I'd probably use some sort of NMR at a high scan frequency (ie higher resolution) and create a 3D map of the geometry of the CD (ie where are the edges and where are the bubbles). Then I'd import that into a AutoCAD and use AUTOLisp to overlay a standard template. Then the AUTOLisp routine could deduce what recognizable filesystems were on it. From there it could convert it into a dd sort of image. This would get more complicated for multi-layer disks.
Note that using AUTOLisp doesn't bode well in the speed deamon department.
But probably faster than factoring a 2048-bit modulus, no? (of course, once you have the contents, you still might have to do that) Not being familiar with NMR, I wonder if it would be equally effective on CD-ROM, CD-R and/or CD-RW. CD-ROM has a solid physical geometry to it, but recordables would be less physically definite, I'd think. Antidote to my ignoragce is welcome.