Indeed, as pinkunicorn said, it’s always the header co-ordinates that are taken into consideration for all geographical statistics.
Strangely enough, while PGC corrects wrong attribution to level 2 administrative units ("states" in the statistics) it believes the CO when it comes to the country. Which leads to several kinds of distortions in statistics, starting with wrong totals for countries (as caches are counted towards the wrong country), wrong extreme caches (NESW), absurd maps (when the area shown on a map is artificially extended in order to show a cache far outside the country which a CO by mistake or deliberate misinformation defined as in that country).
Ideally, PGC should do as for lower administrative levels and override CO’s wrong attributions. I do not know why this is not done, maybe there is an IT related reason for this?
But in the absence of such an automatic correction I would strongly advise against placing a cache or it’s header outside the country to which it is attributed. First, because it creates all the problems described above and secondly because I find this terribly unfair. Of course, you can always put the header of a mystery further north/east/west/south than existing caches, by moving it into a different country (but attributing it wrongly) or moving it into the sea (if applicable), but I do not think that this is fair on the owner of an existing traditional (or other cache) which is really situated on or near an extreme point of that country. So, please, DON’T DO it! (And remember, as long as PGC does not extend the override procedure to the top ("country") level, there can always be another CO placing the header of a mystery even further outside the country, snatching that "honour" from you.