I’m expanding my comment into an answer. Assume $y < x$ and let $z=\ln_y{x}$: then $z>1$ and $z$ is irrational.
The condition is false iff there exist increasing sequences $u_n , v_n$ of integers such that $x^{u_n}=y^{v_n}+o(1)$ (*).
Now, if $u_n,v_n$ are increasing sequences of integers, $x^{u_n}+o(1)=y^{v_n}$ iff $v_n=zu_n+\ln_y\left(1+o(x^{-u_n})\right)$. But $\ln_y(1+u) \sim \ln{y} u$ as $u$ is small, so that $x^{u_n}=y^{v_n}+o(1)$ iff $zu_n-v_n=o(x^{-u_n})$, ie iff $|z-\frac{v_n}{u_n}| = o(x^{-u_n}/u_n)$.
Thus, your property is false for the pair $(x,y)$ iff for every $\eta>0$, there are infinitely many pairs of positive integers $(p,q)$ such that $|z-p/q|< \eta x^{-q}/q$.
This is an extremely close rational approximation: it implies of course that the irrationality measure of $z$ is infinite (thus $z$ is a transcendent number), but it is, in fact, much stronger.
Assume for instance $z=\sum_{n \geq 0}{10^{-a_n}}$, where $a_n$ is an increasing sequence of integers, $a_0=0$, and $a_{n+1}-a_n \rightarrow \infty$ (eg $a_n=n!$ for $n\geq 1$ and $a_0=0$, like Liouville’s number).
Take $v_n=\sum_{k \geq n}{10^{a_n-a_k}}\in 1+10\mathbb{Z}$ and $u_n=10^{a_n}$.
Then $(z-\frac{v_n}{u_n})\sim 10^{-a_{n+1}}$. To have the approximation property for these two sequences and some $x >1$, we would thus need $10^{-a_{n+1}} = o(x^{-10^{a_n}})/10^{a_n}$, that is, after taking logarithms, $a_n-a_{n+1}+(\log_{10}{x})10^{a_n} \rightarrow -\infty$, that is, $a_{n+1}-C10^{a_n} \rightarrow \infty$ for some $C>0$, or in other words $10^{a_n}=O(a_{n+1})$.
This is completely possible, (choose $a_0=0$ and $a_{n+1}=10^{a_n}$, then any $x<10$ will work – that is, be a counter-example to your question with $y=x^{1/z}$); but you can see that even Liouville’s number does not come close to satisfying this property.