Short of elaborate warping with RevisionFX' RE:Flex or equally elaborate re-projection techniques in a 3D program I can't think of much. Such stuff has a way of being a lot more complicated than just cross-fading two images with a bit of corner pinning...
If this was my project I'd probably use SynthEyes to create 3D geometry of the buildings I want to replace then edit the texture maps in my 3D app using the archival photos as one texture map and the new footage as another. You'll have to distort and warp the archival images to match the geometry created from the new shots. This will be much easier if the new footage is shot from pretty much the exact same camera position as the archival photograph.
Another approach would be to do corner pin tracking in Mocha or AE and then use Photoshop to create a document with 2 layers. The bottom layer would be the first frame of the moving video, the second layer would be the archival photo. Now edit and adjust the perspective of the archival photographs in the first frame of video so that it perfectly matches. Turn off the bottom layer and import this adjusted image and apply your corner pin tracking to the adjusted image. If the camera position is very close to the original position you will have minimal corrections to the image. If your camera position is way different, like in your screenshots, then it's gong to take a lot of work with distortion tools.
Both approaches require very careful pre-production planning to pull off. Let me repeat, new footage must be shot from very close to the same camera position as the archival photographs because perspective is a function of camera position. Your moving shots should be slow and not too extreme or you'll end up with perspective shifts that are very difficult to control. You'll then also have to apply optics compensation to your new footage or the archival photos because the lens distortion for the two different lenses will be different.
Yeah, that's a good idea, I'll try that. I was watching some cool videos of people using the perspective warping Photoshop tools that seem promising. It would be great to have that in AE somehow. Or a way to attach pins to the main corners of the building and then attach those pins to nulls in a 3D track so the pins stay attached together through the video and warp itself somehow. Wishful thinking!
Mylenium maybe Projection Man in C4D?