It is not hard to prove a weaker result based using the seminal work of Sims (1961) and Higman (1964).
Proving your result, I fear, will require a deeper understanding of $p$-groups than currently exists. Let $f(n,p)$ be the number of groups of order $p^n$ and let $f_2(n,p)$ be the number of such groups with $\Phi(G)=G'G^p$ central and elementary abelian.
It follows from Sims that $f(n,p)\leqslant p^{cn^3+dn^{5/2}}$ (see [1]) and from Higman that
$f_2(n,p)\geqslant p^{cn^3+en^2}$ where $c=\frac{2}{27}$ and $d$, $e$ are constants.
Note that $d>0$. Higman had $e=-\frac{2}{9}$.
How does this relate to the ratio $A(n)$ in the above question? The groups of exponent dividing $p^2$ (and order $p^n$) are precisely counted by $f_2(n,p)$. Since a group of exponent 2 is elementary, we have $$A(2^n)=\frac{f_2(n,2)-1}{f(n,2)}\geqslant\frac{2^{cn^3+en^2}-1}{2^{cn^3+dn^{5/2}}}\to 0\qquad{\rm as}\ d>0.$$ Taking logarithms of the numerator and denominator above (and ignoring the $-1$), this ratio approaches 1, that is $\lim_{n\to\infty}\frac{cn^3+en^2}{cn^3+dn^{5/2}}=1$. Even if there were a constant $d'$ such that $f(n,p)\leqslant p^{cn^3+d'n^2}$ as Sims conjectured on p. 153. We must have
$e<d'$ as $f_2(n,p)<f(n,p)$ and then $$\frac{f_2(n,2)}{f(n,2)}\geqslant\frac{2^{cn^3+en^2}}{2^{cn^3+d'n^{5/2}}}=2^{(e-d')n^2}\to 0\qquad{\rm as}\ e-d'<0.$$
Perhaps an expert on these error terms can say more.
[1] Simon R. Blackburn, Peter M. Neumann and Geetha Venkataraman,Enumeration of finite groups. Cambridge Tracts in Mathematics, 173. Cambridge University Press, Cambridge, 2007
Length(AllSmallGroups(Size,256,IsAbelian,true));since this is precomputed (seeSmallGroupsInformation(256);). Also, $m=9$ is the maximal $m$ you can do with GAP's Small Groups Library. I don't have any ideas about the proof, though. – Olexandr Konovalov Dec 23 '15 at 13:25