Obviously, Amatus is the expert here.  I did get the opportunity to see
the latest CEREC machine at my study club.  Very impressive.  Amatus
seems to have an excellent experience.  The earlier versions did not do
the occlusion, and I believe he was able to grind these into occlusion
in the mouth.  The newest CEREC is supposedly pretty good on occlusion,
and occlusal adjustment is far easier.
Amatus is obviously quite busy, but I'm sure he'll answer when he can.
He has said in the past that there is a learning curve, but he's never
said there was a problem with marginal fit.  On the demonstrations I've
seen, it seems to be very accurate, and I see no reason that the
marginal adaptation shouldn't be as good or better than a PFM.  After
all, there are fewer materials involved to distort--no impression
material, no die models, no investment models--all of which are
potential causes of distortion.  So if the optical scan is set up right,
it should lead to superior fit.

Steve

Which systems did you look at?

The amount of occlusal adjustment needed is dependent on your trust of the
machine.  You can adjust the settings so that the restoration does not need
any adjustment if you trust the software.

Adhesion is never a problem.  You have to use proper bonding materials in
the proper way.  Fractured porcelain is due to insufficient bulk of material
or lousy occlusion.  Also, you can have a fracture if you leave an air
bubble in the bonding agent.  Contamination is another issue.

The only ones that come off are due to contamination or to faulty silane.