I'm in the market for a new body with the criteria being superior low light capabilities, something my 1DsIII is somewhat lacking, and superior IQ.
Dynamic Range and bit depth also rank in the want column as well. I'm thinking that photomerge will overcome the need for high MP needed for large images.
Noise is NOT primarily determined by pixel size as so many people falsely assume. It is primarily determined by technology and total sensor surface area. When you stitch images what you are basically doing is simulating a larger sensor surface area.
Low ISO noise has been a solved problem for a long time now, and no modern sensor produces noisy low ISO images when properly exposed. People who claim otherwise are zooming to 300% in PS OR using image viewers with horrible scaling algorithms (Apple's Preview is one) OR trying to lift shadows by 3-5 stops. Or all of the above. Suffice it to say, noise is going to be even less of an issue when you stitch multiple frames and thereby simulate a larger sensor.
Dynamic range OTOH is primarily driven by technology and pixel size, and stitching won't help this. Then again, if you're shooting a subject still enough for stitching, you're shooting a subject that's still enough for exposure blending / HDR.
FWIW - I find stitching 3 frames (camera orientation opposite of image orientation) to be relatively easy and practical. Anything more is a chore. Even the lowest Rebel can match a top notch MF film scan with a three frame stitch. It's rare to need or even see prints that big.
I'm thinking that a crop sensor that utilizes the "center sweet spot" of the lens coupled with reduced MP to allow more utilized light would be the answer.
You're not taking technology into account. The current sensors in either format are the lowest noise sensors Canon has produced to date, and noise is being further squashed into oblivion for a given print size by the stitch. Though stitching doesn't help DR, the current sensors are also the highest DR sensors to date, again due to technology.
If DR is truly a concern then 1DX, 5D3, or 6D. If you're doing this for big landscape prints, be aware that the DR gains aren't a great benefit in practice. Generally if your landscape scene exceeds the sensor DR, it REALLY exceeds it, and you need to exposure blend/HDR. In which case any sensor will do.
Don't forget to buy a panoramic head.
You may also want to consider if the shift function on the T/S lenses better serves your needs.