Well correct me if I'm wrong but that statement is false...my s3 runs on at least 2 voltage planes.
The power management on it is also extremely good, batterylife is superb..especially considering the luminous 4.8 inch hd screen.
Also power gating also has worked flawlessly from the first day I got it...able to shut down 3 cores if need be.
You are wrong, I remember you claiming you seeing it run on different frequencies per core, I don't know what program you used to monitor, you might have confused that with core load, and not frequency.
The ARM cores are fed by a single voltage rail from the external PMIC, for it to actually have per-core DVFS it either needs 4 rails on the external PMIC, or for it to have on-chip regulators. In any way the CPUs run on a single frequency plane and you can see that in /sys/devices/system/cpu/cpu0/cpufreq/affected_cpus. Hotplugging the CPUs is a limited and only rough and coarse power management feature and currently it samples every 2 seconds on wether to turn off a core or not, and hotplugging itself is very latency bound by the operating system due to kernel architectures and how it handles migration and creation/destruction of the kthreads, resulting latencies of ~80-150ms for current kernels on consumer devices, compared to actual transition times of dozens of microseconds for which the hardware is capable of.
With newer 3.5+ kernels we now have coupled cpuidle frameworks in the kernel allowing for efficient entry into power-collapse C-states, which affect in most SoC implementations the whole CPU power plane. The hardware itself is not a limitation, but the software is. Precursor to this we could have power-collapse on a core while top (screen on) was running only if that core was the only one running. On the Exynos this is called AFTR/Arm Off Top Running. I'm currently trying to port this myself onto the arch cpuidle driver, but since the system is running on a 3.0 kernel, there is a lot of backporting to be done. The current driver is also dubious in terms of the specified residency times for WFI, being about 1000 times higher than any other architecture out there, and since this differs greatly to the values in the updated driver for the 5250, I think it is a big overlooked mistake in the kernel code.
Did you not ever ask yourself why the device runs so hot in screen-on idle? 40°C CPU temperature at 200MHz while supposedly also clock-gated most of the time, nevermind even hotplugged other cores, is not normal. The phone would have much better screen on battery life if it's power management would be optimized.
