And I'm coding a renderer which takes less time... but this takes also a long time ;-)
Lacking iMovie or any other convenient movie editing software, when crunch-time came on my Brief History contest submission I wrote some. It took a few hours in each of two sessions on two successive days, using a decent Lisp environment and preexisting libraries for getting jpg and png files into memory as arrays of rgb triplets and for writing them out again as pngs.
It helps that its "user interface" was merely a Lisp REPL, and its "documents" are a directory of source images (stills and frame sequences) and a description, essentially, of how to combine stills and clips with transition effects. The editing is not wysiwyg; it's to things like iMovie as LaTeX is to Wordperfect. But that let me focus on the business logic, and to also use an iterative refine, adjust, compile, preview approach to producing the final product. (Change also used it, a bit later.) And the results, you must admit, were of good quality, production-values-wise.
Of course, using my own software also meant being able to write any transition effect I wanted. The zoom box animations are actually transition effects. (The inset map in the Julia morph sequence, OTOH, was done in the fractal renderer itself, using masking and layers.)
Another thing I got to ignore was figuring out things like avi, codecs, and similar nonsense (and sound!). With a png-stream output from the movie-making code and an mp3 generated with Audacity I could just use Virtualdub to put together an avi. Why reinvent the wheels I already had in Audacity and Virtualdub?

(More time-saving efficiency shows up in the source image generation. The ending Mandelbulb spin used actual calculations for 1/7 of one turn of the bulb and looped this, exploiting the bulb's symmetry; the zoom sequence was computed as 50 or so keyframes at 3840-pixel-wide resolution, of which half were used as stills in the zoom-box animation sequence and all were used to interpolate the continuous zoom movie. The second slowest thing was actually the Julia morph, because parameter morphs don't lend themselves to interpolation and the motion blur effect used there is also expensive, while the slowest was the Googol deepzoom still, which I didn't have sitting around already computed. That one took four days to calculate.)
In any event, the movie "editor" program amounted to about a thousand lines of code. I'm not sure what renderer you are coding, but depending on the intended audience/user base and how many features can be supplied with other external tools, a similarly minimalistic approach might be possible.