Yay Pottsey troll posts
We've gone through this several times already
Kona's right. Tegra 2 doesn't completely accelerate the H.264 decode pipeline, so there's some software offloading that use the A9 cores and they don't have neon architecture. Doh!
Performance and battery life suffered, Android and flash did eventually start leveraging the GPU but it's just papering over the cracks, you don't want the GPU decoding video.
(Only the ziilabs stem cell GPGPU is designed to do that and it's yet to be proven for performance/power consumption)
Kal-El, apparently, adds a massive decoder and the full neon SIMD, job done.
Oh Pottsey, these SoCs are all about heterogeneous computing, ideally your main GPU is completely switched off until you load a game. A small, low power architecture GPU is good enough for UI and browser acceleration etc (see OMAP4) Even Tegra 2's GPU is overkill for those tasks.
Regarding 3D gaming performance it looks like Kal-El is roughly about the same as a highly clocked SGX540 or about 50-80% faster than Tegra 2, it's the same architecture with more 1/3 more cores and clocked a bit faster.
(I assume the SGS2/Nexus hit the fps buffer on that benchmark)
I would've liked to see a bigger improvement but nVidia have to sell this chip to OEMs for around 6 months and there won't be any games, in that time frame, that'll require huge horsepower.
The Apple A5 was specifically designed for gaming but skimped a bit on other parts to keep the die size down, i.e. it uses the same encoder from the A4. (don't forget Kal-El is 30% smaller than the A5)
There's a strong chance Apple will keep this performance level for another year too.