> To check that all <agent>s contain the same value use:
> 
> agent != agent  (one of the few times you want to use !=   ....!)
> 
> or maybe in 2.0:
> 
> count(distinct-values(agent) eq 1)
> 

agent != agent is probably going to have O(n^2) performance. It can be
optimized of course but it's not a common construct so that's unlikely.
O(n^2) is the worst case, but the worst case occurs when the data is
correct, so it's also the most likely case.

count(distinct-values(agent)) is likely to have O(n log n) performance.

You can do it in linear time with

not(agent[. != following-sibling::agent[1]])

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