(1) I understood that in the language of first order logic quantification cannot be over sets (there are no predicate variables that can be bound by quantifiers), whereas the language of second-order logic extends the language of first-order logic by allowing quantification of predicate symbols and function symbols (where these symbols may be assigned sets as their extensions http://plato.stanford.edu/entries/logic-higher-order/ ).
(2) I understand that in first order set theory (which uses a first order language) the (first order) variables x,y and z range over sets.
Suppose we have some mathematical theory, T, which concerns some kind of entity. Externally to T, these entities might be able to be understood as sets of sets (but lets suppose that they are not defined as such in the formal theory, T). They would then form the domain of the mathematical theory. Could we, in that case, use FO to quantify over such elements of the domain in this mathematical theory? If so, then it would seem the distinction between FO quantification and SO quantification becomes theory relative in a way which is rather strange. Does this not trivialise the distinction somewhat? While Full SO quantification allows quantification over arbitrary subsets of the domain, certain of those subsets might be elements of the domain of our hypothetical mathematical theory, T, and so in T we might quantify over these elements, treating them as substitutable for FO variables.