As for comparing Gecode with CP Optimizer, the largest difference IBM CP Optimizer is a closed-source commercial system, with paid support. CP Optimizer also has some (as far as I've seen) amazing automatic search heuristics, automatic symmetry breaking, and more advanced scheduling constraints. Gecode on the other hand is open for modifications, has nice parallel search built in, and can be embedded easily. Given a problem where both systems have the constraints, and the search heuristic is fixed, my guess is that they would be mostly equivalent in speed
Also, since you seem to be interested in using Minizinc for modelling, you can use other back-ends than the built in one for solving models.
For some more code examples, see also Håkan Kjellerstrands blog (http://www.hakank.org/constraint_programming_blog/) which, among other things, contains numerous solutions to the xkcd problem suing different systems.
Is there a way to get a minizinc model into cp optimizer?
As for Håkans pages, http://www.hakank.org/common_cp_models/#xkcd has a list of 21 implementations of the xkcd problem.
The IBM CPLEX CP optimizer is best for this kind of stuff IMHO.