Instability of long cylinders

>From time to time on these maillists the subject has come up that
spheres and tori might be better shapes for habitats than long cylinders
because of the latter's tendency to eventually tumble end-for-end.
While on the newsgroup rec.arts.sf.science somebody was talking about
the ultimate size for space habitats given current materials, and I was
mentioning the dimensions of Island 3. This prompted James A. Donald to
say:
"Island 3 is not practical. Being longer than it is
wide, will be unstable. Will flip over, killing
everyone inside."

To which I responded:

"Eh, you're referring to a real-life aspect of physics, but needlessly
hyperbolizing it. It's more like if civilization fell in an O'Neill
cylinder, and the stabilization systems failed, and if you went away and

came back after a few decades, it would be tumbling end-for-end rather
than
on its long axis.

"Frankly, I consider the mirror design of Island 3 to come closer to
making
it impractical than the simple fact that it's longer than it is wide.

"But the point relevant to Knobby's point was that cylinders 4 miles in
diameter are possible with present materials."

He came back to say that "It would tumbling be end for end in about ten
minutes or half an hour" and further went on to say that an above-ground
swimming pool breaking and dumping its water would be sufficient to
destabilize the cylinder.

So what say you guys? A problem if the automatic stabilization system
remains off-line for weeks or months, or a catastrophic flip within the
hour, killing millions?

But ultimately, it occurs that my very first response should have been
that an O'Neill Cylinder(s) habitat would actually have 0 net angular
momentum because it's actually two counter-rotating cylinders. In The
High Frontier, O'Neill only mentioned this as a solution to the problem
of turning the axis of the habitat so as to track the sun, but could it
be that this design solution was also proposed for the second reason of
the instability of long cylinders?

Regards,

Mike Combs