> The first one was eXist. With that database, using a chain of 
> predicates was much faster then using a single predicate with 
> and operators. Obviously, this is a question of using 
> intermediate result sets or not. 

eXist has quite a good reputation - I would be rather suprised if it is
using intermediate result sets to implement a filter expression, however it
is written. I would think it's more likely to be a question of detecting
opportunities for using indexes. But I might be wrong, I don't know it that
well.

Unfortunately it's a truism that optimizers aren't predictable.

Michael Kay
http://www.saxonica.com/