What you now can test is, whether the problem is only with the "interactive capture" or with the screenshot feature (based on Java Robot - see other bug) as such:
print capture(Screen(0))
print capture(Screen(1))
and inspect the images, that are stored somewhere in temp
Since I cannot test on Linux real machines and even not with more than one monitor, I wanted to be sure about the screen setup as Java sees it.
Now I agree with your idea about the wind manager.
There is another bug that goes into a similar direction: /bugs.launchpad .net/sikuli/ +bug/1704302
https:/
What you now can test is, whether the problem is only with the "interactive capture" or with the screenshot feature (based on Java Robot - see other bug) as such:
print capture(Screen(0))
print capture(Screen(1))
and inspect the images, that are stored somewhere in temp