Well the points (in slighty random order) were

- with 'real' framebuffer in main RAM (SDRAM) whole system is slowed down by constant LCD display refresh that fights with any other access to RAM. When they designed 770 someone decided this is too much to bear and they used external LCD controller with its own memory. Same solution is used on PCs, integrated videocards steal RAM cycles from main RAM too, dedicated video cards are preferred for better performance. This solution however makes display updates more complicated as, unlike with PCs, external video RAM is not directly mapped to CPU and cannot be written to easily (and directly). It must be send 'by hand' (you can send specific rectangle of display area).

- OMAP 2 in N8x0 has SRAM good enough for 640x480 which is sadly less than 800x480 our tablets need, so problem is not solved and there is still external chip with its own 800x480 framebuffer. internal 640x480 SRAM is used for additional video plane (why not when we have it) but this does not bring big benefits now as it also needs to be transferred to external chip (together with rest of normal 800x480 framebuffer in SDRAM). I believe there is slight benefit of having this SRAM used now for video and some video scaling or other transformation is done by OMAP display controller but nothing substantial.

- as for 3D - maybe the video plane in SRAM would be a benefit but maybe not, I don't know if the (currently unused) 3D chip can draw to it or must draw to SDRAM and what is performance of both solutions (if both are possible). Maybe it doesn't matter much and this is not the main problem, see below

- you can draw as fast as you can (in 2d or maybe in 3d in future) to internal 'video' RAM (both SRAM and SDRAM) but you won't see it on display until transferred to external chip. This causes problem for fullscreen 800x480 display updates at good FPS rate but is supposedly good enough for 640x480 at ~25 fps so it is not show stopper but ..

- current video system is pretty complex even now so it is not surprise that nobody at Nokia pushed hard for making 3D acceleration working. Even if hardware specs could do it, it is not easy to make it running with current architecture so (IMO) they better spend developer resources elsewhere. Also maybe the licensed 3d solution Nokia could buy would not fit this architecture and customizations would be costly or even impossible. Maybe all this can change if Nokia managers dazzled by iPhone decide to change some priorities but I'm not holding my breath for it :-)

- hopefully with next generation we may fit whole 800x480 (or more) to directly mapped video ram and ditch external chip making the architecture easier to support features like 3D, video out, etc

Not sure if all this gives you some arguments in your iphone vs Maemo SDK discussion though :-)