I have a K 28/3.5 Shift lens and have used it a reasonable amount (its not going anywhere nor looking for a new home). I have also used it in ways that are somewhat off nominal - essentially producing a 4x5 (stitched) resulting image, based on how the lens operates (across the clock points as the lens rotates) Yes, the 28mm is a bit narrow for use on an APS-C cropped sensor, so you are not able to receive its true capability. It is also a shift only and not a tilt/shift lens, so some may view this as a limitation.
I really wanted this for the shift capability and was not that interested in tilt. I use the shift for stitched panoramas. One aspect that folks miss, is the combination of both horizontal and vertical stitching into a single image, rather than just a single axis.
I have thought through this this problem a number of times and there are really no good solutions for a number of reasons.....
- Lens design - For both the shift and tilt operations - and especially the combined operations, the lens needs to be designed optically to have superior edge and corners, essentially across the entire lens, rather than just at the center. You are shifting/tilting the lens off its optical center so that it is positioned over the sensor such that the lens' edges and corners are closer to the sensor's center (then normal), and thus the projected image from the lens on to the sensor needs to have a high image quality. Also, as the lens is shifted over the sensor, the projected image circle needs to be substantially larger, so that the sensor will be covered at all shift points and extremes.
- Image on the Sensor - the K 28 shift is designed for film. In fact, I somewhat doubt that you can design an perfect T/S lens for digital sensors. What am I talking about. Film as a sensor had an chemical emulsion surface as a photo receptor (essentially its a pretty nice flat surface receiving the light). Digital sensors, have the photo receptors down at the bottom of pits or dwells (think of dimples on a golf ball), with potentially a micro lens cap over the top. What this does is to essentially optimize the light rays coming into the dwell from a perpendicular angle - i.e., straight from above. So light rays coming in from the side, will be coming in at an angle - which may be deflected away by the depth of the dwell or the micro lens caps of the photo receptors on the digital sensor. So, what does this mean - when you have the lens shifted over to one side, the light coming through the lens edge (which is now over the center of the digital sensor) needs to fall on and illuminate the far edge of the digital sensor (thus coming in at an angle). Essentially the quality of the light making it over to the far side of the sensor is less than optimal.
So, in putting all of this together, for a reasonably wide Tilt/Shift lens, you are going to need an optical design that is pretty perfect across the entire projection circle of the lens on to the sensor. That means very low distortion. That brings us to the 24 to 35mm focal length range (based on the 135 format) that is used for Tilt/Shift lens. This focal length range is the bottom (or widest) you can really go with out having to accept large amounts of distortion. Going wider, you start to bring in ever greater amounts of distortion (especially across the edges and corners).
How is this distortion going to affect the resulting combined image when stitched. Well, let's think about this for a minute. Let's keep it simple - a 3 frame stitch, 1 normal image (lens center on the sensor's center), and 2 shifted images one to the right and one to the left. The two shifted images are going to have distortion to the left on the left of center shift, and distortion to the right on the right of center shift. When you take and stitch or combine these images together, you can have a mess.
For Tilt/Shift lenses using both tilting and shifting, you are essentially going to be taking a single image, where the optics in the lens does all of the work. You are going to possibly need a larger image circle to cover the sensor, and you are going to need to have a superior optical lens design across the entire lens because of the off center positioning potential of the lens over the sensor. You are also going to need as little distortion from the lens as possible (again that 24 to 25mm focal length comes into play), as you do not want to add in edge or corner distortion to your perspective correcting lens.
So in the 645 medium format, the 35mm focal length lens equates to something like a 14mm lens on an APS-C sensor format. Therefore the solution is a larger sensor format, so that you are able to maintain a reasonable focal length (24-28mm and longer) which controls the distortion naturally, go to a larger lens surface to work with (yes, more expensive glass), along with a longer registration distance in which to design the mechanical tilt and shifting mechanisms.
Another solution is a view camera, where tilting and shifting are essentially built into the camera's mechanical design.
The last solution that I can come up with is a set of bellows that can be attached to an FF or APS-C body with a lens on the other end. You will need a set of bellows that are capable of both tilting and shifting - they are available. However, by the very nature of the bellows, you are adding focal length, great for macro, not so good for wide angle landscapes or architecture.