Notices

I investigated the temperature thing a little bit further and I made these changes:

On vanilla kernels the temperature raw value is still being used. On power kernels I use /sys/class/power_supply/bq27200-0/temp as a temerature source, which should be more accurate. Unfortunately this is not possible on vanilla kernels.

Very briefly - I am seriously impressed with the improvements this simple app is making to my battery life (CPU settings 125-500, Smart Reflex disabled).

I had tried the power users kernel (Ideal settings) linked by Flandry, and was surprised and disappointed to see my battery life worsen noticeably with that.

I'll keep an eye on Battery Graph for the next couple of days and see how it goes.

Please convert temperature to celsius like this

Battery Graph showed higher consumption using Ideal settings?

FYI the general consensus about 125 MHz is that it actually reduces battery life with well-behaved applications due to the slower race-to-idle (presumably why Nokia didn't use it).

The best explanation that fits with both of your observations is that you have something installed and running that doesn't let the processor sleep much (see e.g. http://talk.maemo.org/showthread.php?t=29284). Have you got some kind of CPU load monitor running to check for the culprit?

OK thanks. That would mean:
Can somebody confirm that this formula is correct and that real/raw is linear?

Is this app equivalent to Gnome's CPU Scaling monitor?

How does CPUFreqUI do it? It seemed to give a real temp. Can't tell if accurate.

Also, I left it there, idle, at room temperature for an hour. At 28 ambient, it shows 47 (raw). By formula, it should say 27 ambient, which is likely true, I get 27.8-28 on all sensors in here. Also, it jumped to 48 fairly quickly after opening app, so it was likely 47-48 for 27.8-28 degrees.

I confirm this formula

Flandry

Regularly checking Conky (if that's reliable enough ?) showed no rogue processes running.

No placebo either - I was surprised (given how happy many people have been with power kernel) but both Battery Eye and the DCEW showing output of lshal pointed to my battery dropping off a cliff on Ideal. No reboots though.

And my battery life on day 2 of QCPUFreq alone is noticeably, actually measurably better ( 68% now after 12 hours use, as opposed to 40 odd percent on stock and 30 odd percent on ideal ) .

I can't explain it, but there.

Overall my little N900 has been exceptionally well-behaved on both software and hardware fronts. I'm quite content to sit back and enjoy my newfound lease on battery life for now

its a wonderful effort no doubt. but id like to say tht if the settings are reverted on every call...thn it makes it a little unusable.
coz u cnt set it again and again after every call!