I realize that it will be a bit more work than I'd like to present comparisons and I would need a flatbed scanner too in there. So I just say what the likely result of such test would be based on my own informal tests this weekend :-)
I change my mind and think 5Dmk2 provides enough density at 1:1, which is close to 4000 ppi. Due to lens and f/8 for enough dof to manage focus resolution is a little bit worse than a 4000 ppi film scanner, but probably significantly better than any flatbed. I did 6000 ppi with the 7D but I doubt it will show on even large prints, the grain is probably adequately resolved with 5D. This reduces the stitching work. I would not want to go lower than 5D density though.
Concerning DR single shot is worse than multipass on good film scanners, but probably better than single pass flatbed. With 3 shot 2 stop spaced HDR (one ettr shot and two brighter for reducing shadow noise) it is better then the multipass scan. Probably single shot is good enough when no shadows are pushed in post, but on dark transparencies you will want HDR since there is detail to find there if pushed in post.
Getting predictable and satisfactory results from HDR software proved a bit messy, but I succeeded with SNS-HDR with neutral preset and reducing brightness slider to avoid highlight clipping. I did stitching in Hugin, outputing one exposure at a time to merge in external software. Hugin's own HDR is too poor quality.
Using this technique one will probably get slightly better quality than a 3200 ppi medium format scanner for medium format transparencies. I think it is certainly good enough for fine art prints.
Use a sharp lens with good corner performance and low chromatic abberations, use bracket mode in camera and live view to minimize vibration, and to make sure no alignment for HDR is needed since that can reduce sharpness.