At NAB I noticed that in some recording modes the encoder is unable to keep up, e.g. 2K with overscan at more than 23.976. I understand that scene complexity has an impact.

Has / is any of this changed?
Are there any rough guidelines as to what is achievable?
Intel keep bringing faster silicon to market, can we upgrade the CPU in the camera?

What did you see that wasn't keeping up?

Was it the SI camera full body or the mini/laptop...

What was the indication it wasn't keeping up?

Yes, CPU's are planned on being upgradeable, so you will have the ability to add faster processors and chipsets from Intel.

As far as not keeping up though, the SI-2K can do full 2K/24P without "falling behind" right now. But you are right, scene complexity does stress the CPU more, but at the default settings 2K/24P is not an issue for the SI-2K. If you start tagging on some of the additional eye-candy and bells-and-whistles that we have in the GUI though beyond the default settings, the encoder can fall behind, but that's what the RAM buffer is for (so i.e., instead of unlimited recording time, you might only have 4 minutes or something like that).

But yes, as noted, with new Intel boards, you can swap those in and get faster performance, so the camera can "grow" with your needs and any new software enhancements/features we come out with in the future.