Ah, I see. I thought you weren't refering to the EVi numbers.
What I've noticed with them (but just impressions, no serious testing), instead, is they seem to stay relatively constant for a given angle as long as the camera is on (or perhaps as long as the batteries aren't disconnected).
I'll try your test, too... I already have a turn table in the form of a piano seat

EDIT:
Wow, you're right. They're both influenced by magnets and change with rotation around the vertical axis... although, as you say, the effects of magnets are small enough to make the conclusion that it's detecting Earth's field dubious. There's also the risk that, if the ground isn't perfectly plumb (I know my place's floor isn't!), that's the effect we're seeing...?

What's for sure is that, if we really only have EVi A and B, then this reactivity to the Earth's field (or whatever it is) will be a hindrance, rather than something useful.
We're already measuring two things (tilt arund two axes)... There's no way we can separately and reliably measure
three things with just two numbers

ATTACHMENT:
I have, too, turned the camera 360 degrees around the vertical (ground-sky) axis in 30 degrees step. I've measured _GetISLensEViAaxis() and _GetISLensEViBaxis() using the integration I described earlier (and which is in my code).
I'd say that the linearity of the A axis is merely due to the integration (the value is still increasing as I write, with the camera still...); as for the B axis, it shows "something", but the changes in numbers are really too tiny to say anything, in my opinion...
Although I'm sure I noticed some bigger changes by eye, earlier. Bah. Anyway, I made sure the surface I did this on was relatively plumb, using a spirit level.
(And, no, I don't know why one graph is smaller than the other - ask Excel)