The CPU usage isn't that bad compared to some of the other things we did in the demo - with some help from interrupts it could be done with maybe 20% of CPU. There is also a much easier ~500 colour variant which doesn't take any CPU time at all once set up.
I think the real reason it wasn't discovered earlier is that most CGA PCs were not connected to composite monitors or TVs (people who could afford the big expensive IBM machine could generally afford a dedicated digital monitor as well). A few games used 160x200x16 composite but even those generally had modes for RGBI monitors as well (which wouldn't work so well with the 500/1K colour modes, though I guess there are the dithering characters 0xb0 and 0xb1 which might have worked). These +HRES modes also suffer from CGA snow, which might have been a deal-breaker.
Cool, thanks for the details!
Well they were able to use the leftover CPU slop to transform, shade, and rasterize a crude 3D model in real time, so it couldn't have been all that bad...
Same thing for all the things you see Commodore 64 demos do... by the time you're creating the awesome graphical effect there's often not a lot left over for the game itself. (Though there are some interesting exceptions... there appear to be some surprisingly high-quality side-scrolling platformers now based on "bad lines", which are both explained and then the platformers shown at https://www.youtube.com/watch?feature=player_detailpage&v=fe... . Though the entire presentation is fascinating and shows a few other demoscene effects in real programs.)