Let $\sigma(n)$ denote the sum of divisors of the positive integer $n$.
Using Sage Cell Server, I was able to get the following odd numbers $n < 5000$ satisfying $\gcd(n, \sigma(n)) > \sqrt{n}$:
$$117 = {{3}^2}\cdot{13}$$ $$135 = {{3}^3}\cdot{5}$$ $$585 = {{3}^2}\cdot{5}\cdot{13}$$ $$775 = {{5}^2}\cdot{31}$$ $$819 = {{3}^2}\cdot{7}\cdot{13}$$ $$891 = {{3}^4}\cdot{11}$$ $$1287 = {{3}^2}\cdot{11}\cdot{13}$$ $$1305 = {{3}^2}\cdot{5}\cdot{29}$$ $$1485 = {{3}^3}\cdot{5}\cdot{11}$$ $$1989 = {{3}^2}\cdot{13}\cdot{17}$$ $$2295 = {{3}^3}\cdot{5}\cdot{17}$$ $$2793 = {3}\cdot{{7}^2}\cdot{19}$$ $$3515 = {5}\cdot{19}\cdot{37}$$ $$4095 = {{3}^2}\cdot{5}\cdot{7}\cdot{13}$$ $$4455 = {{3}^4}\cdot{5}\cdot{11}$$ $$4655 = {5}\cdot{{7}^2}\cdot{19}$$
Is this surprising? Or is there an underlying explanation for this phenomenon?
Updated Question (Added October 30, 2018)
Under what suitable conditions does it follow that $3 \mid n$, if $n$ is an odd number satisfying $\gcd(n,\sigma(n)) > \sqrt{n}$?
Note that it follows from the inequality $\gcd(n,\sigma(n)) > \sqrt{n}$ that $n$ must be composite.
Claim : There are infinitely many odd numbers $n$ such that $$3\not\mid n\ \ \qquad\text{and}\ \ \qquad \gcd(n,\sigma(n))\gt\sqrt n$$
Proof :
For $$n=p^{2a}\sigma(p^{2a})$$ where $p$ is a prime larger than $3$ and $a$ is a positive integer such that $(p,a)\not\equiv (1,1)\pmod 3$, we have $$2\not\mid n,\qquad\qquad 3\not\mid n$$ and $$\gcd(n,\sigma(n))=\gcd(p^{2a}\sigma(p^{2a}),\sigma(p^{2a})\sigma(\sigma(p^{2a})))\ge \sigma(p^{2a})\gt \sqrt{p^{2a}\sigma(p^{2a})}=\sqrt n\ .\quad\blacksquare$$
Some small examples :
$n=5^{2}\sigma(5^{2})=5^2\cdot 31$ which you've already shown.
$n=5^{4}\sigma(5^{4})=5^4\cdot 11\cdot 71$
$n=5^{6}\sigma(5^{6})=5^6\cdot 19531$
$n=7^4\sigma(7^4)=7^4\cdot 2801$
$n=7^6\sigma(7^6)=7^6\cdot 29\cdot 4733$
$n=7^{10}\sigma(7^{10})=7^{10}\cdot 1123\cdot 293459$
$n=11^2\sigma(11^2)=7\cdot 11^2\cdot 19$
$n=11^4\sigma(11^4)=5\cdot 11^4\cdot 3221$
$n=11^6\sigma(11^6)=11^6\cdot 43\cdot 45319$
Added :
For $p\equiv 1\pmod 3$, we have $$\sigma(p^{2a})=1+p+\cdots +p^{2a}\equiv 1+1+\cdots +1\equiv 2a+1\not\equiv 0\pmod 3$$ for $a\not\equiv 1\pmod 3$.
For $p\equiv -1\pmod 3$, we have $$\sigma(p^{2a})=1+p+\cdots +p^{2a}=1-1+1-1+1-\cdots -1+1\equiv 1\pmod 3$$