My mates galaxy ace seemed to return a value based on G so on a flat surface screen facing up Z was -1.0 (which makes full sense). My nokia lumia 710 also does this but is a windows phone of course. My old ZTE blade returned a value in ms-2 which is what acceleration is really measured in. On a flat surface screen up it returned -9.72 which I think means it was out of calibration as 1G is 9.81.
I think in G you can assume that no-ones going to break the -3 to 3G limit which is greater than some of the cheaper sensors used in phones, Thats 29.43 in ms-2
If you keep an Android uplifted onto its side, and continuously calibrate the G-sensors, it's possible that you can force the X and Y accelerometer to exceed the float value: 9.8
For example, X default value is 0.0.
0.0 > Lift upright > 9.72 > Calibrate > While calibrating, shake the phone rigoriously > you get X := -27.83
Source: I have a shitty phone I borrowed from school, and played its calibrations. Quite happy with it, as an expensive toy.
EDIT: I reset the values back to 0, and try this again, so as to take a snapshot for proof. I'm now unable to recreate the values. Please ignore the above. I'll re-edit this post and try again tomorrow.
EDIT 2: Forget it, quickly mashed this up and got this:
Note that this is when I put my phone on the table, laying still. So, play around a bit with the G-sensor calibration, and you'll see how fun it is to tinker around.
EDIT 3: Ok, I got one that even exceeded my initial record for max G-force:
Edited by tom_mai78101, 13 September 2012 - 10:28 AM.