There is a long-held belief that optimization is best left until never. It is generally believed that optimization is a tough challenge and not a rewarding one. So news that simply recompiling Android 4 makes it run twice as fast is stunning.
Linaro is a not-for profit organization formed to improve the quality of Linux on ARM based hardware with members such as ARM, Samsung, TI, IBM etc.
Linaro has been working for some time on improving the standard GCC (GNU Compiler Collection). The latest Android Ice Cream Sandwich 4.0.1 was recompiled using the modified tool chain and was demonstrated running on a Panda board next to an unmodified system running on identical hardware - speed improvements are obvious, see the video below.
Details are sketchy, but apparently the tools include the ability to manually or automatically parallelize tasks across multiple cores. Modifications had to be made to the original source code to allow it to be built using the GCC chain and to allow optimization to be switched on.
Both Google and Cyanogen are reported to be incorporating the improvements into their own software, so we could see Android running faster on production phones rather than just on specialist boards in the future.