Ok, here are my results trying the slideshow-with-ken-burns thing with Cinelerra:
(Note: used the CV on Ubuntu Gutsy) 1. Images are imported as single frames. You need to zoom into the timeline to an insanely low level, then stretch the image a bit, zoom out, stretch again, zoom out again etc. etc. until you have the image shown for about 10-20 seconds. This is already a waste of time. Default should be 1 second. 2. The default compositor settings are such that a 100% crop, with the size of the output video, of the original full-res image is taken. This means that for each newly imported image, you can see only a very small portion of it. Therefore, you have to tweak the camera settings in the compositor before you see something that is closer to what you wanted originally. Again, this is a waste of time. Default should be full image. The general logic of the Compositor is programmer-friendly but impractical to use. 3. Fading between images needs to be performed by setting Bezier curve points for each track. Not as cumbersome as (1) and (2), but defining each transition this way is also wasting a lot of time. 4. Setting keyframes works (sort of) but is invisible and impossible to edit afterwards. Furthermore, keypoints cannot be set to the last frame of a strip, the editor doesn't allow this but shows an empty frame. 5. Rendering is an error-prone and unpredictable process. Worst what I have ever seen. So theoretically, Cinelerra CAN do what I wanted, but it doesn't support a reasonable work flow. While I managed to achieve about 50% Productivity with Blender (compared what could be done with Kdenlive once the linear transformation would be available), it is near 1% with Cinelerra. <sarcasm> May be it is not that obvious that people with day time jobs want to get stuff done in finite time. </sarcasm> Sorry, yet another failure.
