I looked into your measurements.

1. Your power consumption is more than my with lesser VDD1/2 besides 500MHz. Your battery provides more voltage (my stays about 3860), so - I guess, the VDD values (20, 25 and 30) are not very accurate and that may be a reason why somebody has a good OC but somebody doesn't - all VDD values are too inaccurate. Note: vdd1_opp & vdd2_opp are the number of element from array of voltages (like XLV is: "30 20 30 33 38 38 45 45 48 48 54 54 60 67 72").

2. Do you sure you did measure with locked N900 (screen!) for low frequencies - your N900 spends significantly more energy than my N900 even with your lower VDD1/2? (824mA * 4094mV versus my 400mA * 3860mV). Or - did you change some 'ondemand' parameters like sampling_rate?

3. The frequency and power consumption ratios clearly points that you have big advantage in 500MHz. After that (850MHz) the frequency increase by 1.7 is accompanied by energy increase by 1.6, so the energy per clock is pretty close (no advantage). But frequency increase from 125MHz to 500MHz (by 4) collects only 1.4 more energy and each CPU clock spends almost 2.8 times LESS energy on 500MHz (1.4 * 2.8 == 4)

So, if you don't run poorly written endless applications you definitely save more energy on frequencies 500MHz and more. Happy you!