I didn’t. But I’m curios too. The whole matter is more and more central to my job’s needs.
If I got it right, Agile is a bunch of well optimized settings: something that, with practice, many of us found anyway.
The extra speed gain in E-cycles, afaik, comes from a modified sampling algorithm. But actually once I had the chance to try it i didn’t experience that big time savings. Imho, in the presets ecycles comes with, the real speed gain still was the biased tricks that are already in vanilla cycles (ao bounces being the most important). I can’t tell about kcycles, but I wouldn’t be susprised if the situation is similar.
What actually would make a big difference (at least in some of my projects) is the ability to render subsequential frames without reloading the whole scene, in animations where only the camera moves. It would be a game changer for scenes where the sync process can take more than a minute!
Anyway, for animations the real deal is to have a good temporal denoising. I dream of it being added in the sequencer somehow. In the past I also discussed about having a Vapoursynth addon, a sort of Avisynth but python-compatible.
A few words for SID. I used it but it comes with a downside: I have to save heavy EXRs with a lot of information for them to be used or tweaked later again. As far as I know the newest Oidn implementation does a really better job than before: in a quick test I made it seems that it makes multipass denoising not so useful anymore.