I counted  KKbar pairs in the  Xu system on an  unbiased event sample,
obtained   directly  from   EvtGen.   My   apparent  problem   in  the
normalization turned out not to be a problem at all, but just physics.
The rate for ssbar popping  in the nonresonant component in the hybrid
MC is different from the nonresonant MC.

Here are the numbers: 

   ssbar in %      Dedicated MC       Hybrid MC
   B+ resonant      10.2               10.7
   B+ nonresonant    7.3               12.0

   B0 resonant       0.2                0.1
   B0 nonresonant    7.1               11.3

"Dedicated" means that it's just the resonant (nonresonant) component,
no mixture  between the two. If  I run the  "dedicated" nonresonant MC
with the  same reweighting (for  the non-resonant part) as  the hybrid
MC, the  dedicated and hybrid  MC have ssbar popping  rates compatible
within the stat. error.

The resonant numbers agree with Daniele's DECAY.DEC calculation within
the stat. error.