Yes, it is true.

For a function $h(x)$ we denote by $LS(h)$, $LI(h)$ the functions defined as $$LS(h)(x)=\max(h(x),\limsup_{y\to x} h(y)),\\ LI(h)(x)=\min(h(x),\liminf_{y\to x} h(y)).$$

Then $$Og=LS(g)-LI(g).$$

Denote $g(x)=Of(x)$. Note that $g=LS(g)$, i.e. $$g(x)\geqslant \limsup_{y\to x} g(y).$$ Indeed, for any $d>0$ and any $a$ such that $|a-x|<d$, we have $\sup_{y,z\in B_d(x)} |f(y)-f(z)|\geqslant g(a)$, taking limsup in $a\to x$ we get $\limsup_{a\to x} g(a)\leqslant \sup_{y,z\in B_d(x)} |f(y)-f(z)|$, now take limit in $d\to 0$.

So we have $OOf=Og=g-LI(g)$, $OOOf=O(Og)=Og-LI(Og)$. Thus your relation $OOf=OOOf$ rewrites as $LI(Og)=0$. In other words, we should prove that for any $\varepsilon>0$, any $x\in [0,1]$ and and $d>0$ there exists $y\in (x-d,x+d)$ such that $Og(y)<\varepsilon$. Denote $s=\inf_{(x-d,x+d)} g$ and choose $y\in (x-d,x+d)$ such that $g(y)<s+\varepsilon$. But $LI(g)(y)\geqslant s$ and we conclude that $Og(y)=g(y)-LI(g)(y)<\varepsilon$ as desired.

Oeuvres Choisies), then generalized by Henry Blumberg 1 2 3 to cases in which the oscillation is computed by allowing for the neglect of countable, measure zero, and first category sets. Although the result appears in a few books (e.g. Hahn's 1921 treatise on real functions) and papers,(continued)$\endgroup$ – Dave L Renfro Dec 15 '18 at 12:44