Neil gave me the idea of testing to see how much differential flexure I have in my side-by-side setup and I got around to doing that last night when we had partly cloudy skies such that I didn't want to try real imaging, not knowing how long it would last. My AT111EDT (with QSI 583 camera) and Orion ST80 (with Meade DSI camera) are mounted on a 3/8" aluminum plate, about 8" apart. Recently I also added an aluminum bar across the tops of the rings to couple the 2 scopes there in addition to the bottom. The AT111 has a very solid focuser and I locked it down for this test. The ST80 (of course) has a horrible, sloppy focuser, so I added a block that attaches the draw tube (right in front of the camera) to the dovetail. It looks and feels like a very solid system.
To test this system I selected 5 stars, Alkaid, Regulus, Vega, Castor and Spica. Alkaid was near the zenith and the others are (very roughly) at 20 to 40 degrees elevation and 90 degrees apart in azimuth. I chose bright stars to make it easy and certain to locate the same star in both cameras. At each star I captured a frame from each camera very close to the same time. I then used Maxim DL's tool to find the centroid of the star in each image and recorded that information. After doing this for all 5 stars I created a spreadsheet where each measurement is converted from pixels to millimeters from the edges of the imaging chip (because the cameras have different size pixels). Then I could find the offset between the two images by just subtracting the X or Y location in mm of one image from the other. This offset, in itself, is not particularly important, except that it has to be small enough that you can actually see the same star with both cameras!
Using the zenith star as a reference, I then calculated the difference in X and Y for each of the other stars. This represents the difference (presumably due to flexure) in offsets between the cameras (in mm) at each position. These numbers were then converted back to pixels and expressed as pixels per degree (dividing the offset by (90 - altitude) and pixels per minute of exposure.
The cameras were both oriented such that X is DEC and Y is RA. In RA, the test shows that a 30 minute exposure would have only about half a pixel error from flexure. Unfortunately, DEC is a different story, with about 10 times as much error. This would probably be OK for exposures up to 5 minutes, but that's not enough for narrowband imaging - or if I switch to a longer focal length scope.
What is missing in this analysis is where the flexure changes most quickly. Intuitively, it seems that it would be when passing through "vertical", so I might be able to get much better results than the test would suggest by limiting the time frame over which I image. But this is just a hunch.
The other test I did last night was to confirm the flatness and tilt of my imaging scope. I recently acquired a William Optics "Flat 4" reducer/flattener and was quite impressed with the appearance of stars, but analysis with CCD Inspector showed that there was a 9% "tilt". This number represents the variation in FWHM across the image. On inspecting the attachment of the Flat 4 to the focuser I found that tightening the thumbscrew visibly tilted the Flat 4. This might have been corrected by simpler methods, but I really didn't want to rely on a crummy little thumbscrew to hold the rather heavy and expensive camera, so I made an adapter that to allow the Flat 4 to be directly threaded on the end of the draw tube. In this configuration the tilt went down to 2%.
I acquired images for analysis by selecting a field of similar stars (no really bright ones in or near the field) near the pole. Being near the pole and using a short exposure minimizes tracking errors.
CCD Inspector also reports the average "aspect ratio" for images. I'm not sure exactly how they do it. For example, a great many stars are binaries that appear as a single, distorted star. Can it detect and ignore these stars? I don't know. When I randomly sample stars in MDL the ratio varies from 0 to maybe 14% (0 is perfectly round). In the current configuration I can't see any "non-roundness" anywhere, while I could before correcting the tilt. CCD Inspector reports aspect ratios of 8 to 16%. I don't know how that compares with others, but I'm very happy with the appearance of stars. That is, in these short exposures. Tracking errors are another matter...