The XQuery semantics are (largely) functional.  Does this open the door
for optimizations that have been developed in the domain of compilers
for functional programming languages?

One such ''functional gem'' could be memoization of expression values.
In the simple XQuery expression below

       for $x in (1,2,2) 
         for $y in (3,4,3)
           return e($x,$y)

the nested FLWOR will lead to a 9-fold evaluation of expression e (in
which variables $x and $y occur free).  Alas, only four distinct
variable binding pairs will be generated by the nested FLWOR.  Can we
evaluate (the potentially costly) expression e only 4 times and then
later ''compensate'' for this optimization to ensure that the resulting
sequence will be of length 9?

Of course, one challenge is the ''(largely)'' above... If e contains
side-effecting constructors, for example, memoization (re-using the
results of e) might be unsound or at least complicated to achieve.

Interestingly, memoization *should* go well together with the
current formulation of the XQuery Update Draft, if the compensation
action is performed in a sensible fashion.

Are there more techniques we could adopt from the FP languages/compilers
domain?  Judging by his presentation's title, I should talk to Kris
Rose. ;-)