Another approach is to compare memory usage of single processes, as this is what thumb2 is about, yep?

We don't care how much of it is swapped or how many widget's one have in desktop - we have list of thumb2 compiled programs available in thumbs repos, and what interest us, is their memory footprint compared to stock (non-thumb) equivalents.

At least, this sound sane for me, or we will drown into hell of non-comparable results. I know that amount of things I've included in data provided via post 1 exceed it - I did it on request, and was kinda sleepy

So, unless someone have points I've missed, proper scientific approach would be to take list of thumb2 compiled programs (from thumb2 repo) and compare memory used (absolute numbers), when they're in use, with non-thumb2 ones. This way, no need for 2 N900's per test's contributor.

/Estel