Behavior of genus function on a 4-manifold for sums
Sometimes the function $G$ can be constantly 0: consider the class $x = [S^2\times\{p\}]$ in $H_2(S^2\times F)$, where $F$ is a surface. Then $G(nx)$ can be realised by an embedded sphere for all $n$: just pick $n$ distinct points $p_1,\dots,p_n$ in $F$, and tube $S^2\times\{p_i\}$ to $S^2\times\{p_{i+1}\}$ (using pairwise disjoint tubes).
As for the existence of a limit, to me this is a lot less clear. Certainly something is known when $b^+(X) > 1$ and some Seiberg–Witten invariant of $X$ does not vanish, at least in the case when $x\cdot x > 0$. Then there is the adjunction inequality (Kronheimer–Mrowka), telling you that (for some second cohomology class $K$, corresponding to a non-vanishing SW invariant) $$ 2G(nx) - 2 \ge |\langle K, x\rangle| + n^2x\cdot x. $$ The right-hand side of the inequality grows quadratically, so $G(nx)$ goes to $\infty$.
I'd be very curious to know of "interesting" behaviours of the function $n \mapsto G(nx)$ (e.g. non-monotonicity, frequent non-monotonicity, eventual constant non-zero behaviour, periodicity/aperiodicity).
In the case that $x\cdot x \neq 0$, topological methods based on the G-signature show that the genus goes to infinity more or less quadratically in $n$. (I'll be more specific below.) This goes back to Rochlin (Two-dimensional submanifolds of four-dimensional manifolds) and Hsiang-Szczarba (On embedding surfaces in 4-manifolds) in the 1970s.
Following Rochlin's version (since I don't have the other at hand): if a homology class $\xi$ is divisible by $h$, an odd prime power, then $$ g \geq \left|\frac{(h^2-1)(\xi \cdot \xi)- \sigma(X)}{4 h^2}\right| - \frac{b_2(X)}{2}. $$ Writing $\xi = h \alpha$ we see that the right side grows quadratically in such $h$. (Generally this grows as the square of the largest prime power dividing $n$ where $\xi = n \alpha$; presumably the growth rate of that quantity in $n$ is known.)
By looking in a neighborhood (and sticking to prime powers), you can see that you'd expect quadratic growth, but the estimate above looks off by a factor of two. For instance, when it holds, the adjunction formula (as quoted by Marco above) gives a bound that is roughly twice the G-signature bound.
Work of Strle (Bounds on genus and geometric intersections from cylindrical end moduli spaces) gives stronger results for surfaces of positive self-intersection in the case that $b_2^+(X) =1$, without the assumption of non-vanishing Seiberg-Witten invariants. See also recent work of Konno (Bounds on genus and configurations of embedded surfaces in 4-manifolds).
Finally, in the case of self-intersection $0$, the growth is at most linear (and possibly $0$, as Marco notes). This follows by tubing together parallel copies of a given surface.