Great idea, thanks for kicking it off!
I just finished evaluating Combustion (used it on an Athlon 1200/266 with FireGL2, 512Megs RAM, Internal ATI Drive).
My background in compositing is strictly Flame/Inferno, used them since 1994 on "old" Onyx's, up to the "blazing" OnyxIR2s, HD capability, etc.
While I love the individual tools (ok, not the warper), I have had long standing serious issues with implementation, integration, etc.
I have head that C* was a great "little" tool, but not made for client sessions. Well, I do not concur! Yes, clients want Inferno, and yes, interaction is faster, BUT!!!!
In almost every way and every category (esepcially with the upcoming 2.0), Commotion is a far, far, far, far superior compositing package. It addresses nearly every shortcoming of Inferno with very little tradeoff. I went back and figured out how I would have built projects I finished in recent months, and in almost every case C*s superior workflow would have more than compensated for slower interaction and rendering (I personally think given the fastest NT and a complex job with filters C* would be very close to Inferno). One might say that having to load clips into RAM for playback is a limiting factor, but think about it; loading a single-layered render is quite fast, so no great loss here. Inferno cannot render-to-ram, so testing renders involves multiple renders (i.e. every time you wish to see a change), rather than just keeping the clip playing and seeing changes on the fly.
So far I have about 6 pages of advantages of C* over I*, and maybe three line-items of disadvantages;
Slower interaction (gap closing)
No bicubic surfaces (third-party options?)
No import of objs.
No streaming video straight from disk.
No warper (third-party option?)
No schematic (addressed in 2.0)
No expressions (?)
No real-time playback of garbage mask splines
Verdict: I would choose to do almost any project in C* over I* (on a decked out Athlon IMHO), unless a client specifically wanted to be in on it and demanded it. I would also try to get clients who trusted my judgement to let me work C*. Imagine their astonishment when we approach a point where I need to separate front and matte layers, apply and modify multiple filters and assign and change regions of affect, combine keying algorithms, etc. and did not have to go through the torturous steps we are used to in I*
I once made a side-by-side comparison of a complicated tracking/keying project to compare what steps should have been necessary vs. the steps forced on me by I*. The verdict; 6 steps took 54 steps. I will try to re-create to see how many would have been needed in C*. Let's get that number to six and the box we workd on won't matter anymore.