Does it make any sense if you put zero order hold after a step input (at any sampling period). I think ZOH does exact reconstruction of a step function regardless of the sampling period.
In the following problem the input is a step function and the ZOH comes after this. Shouldn't we get the same response as if it were a purely continuous (not sampled at all) system? discretize a function using $z$-transform
$h_{\text{ZOH}}(t)=\text{rect}(\frac{t-T/2}{T})$ is used for reconstruction of a sampled signal which is represented by $s[n]=\sum s(nT)\delta(t-nT)$. So the output is $h*s=\sum s(nT)\text{rect}(\frac{t-T/2-nT}{T})$.
Now if we follow two appropriately scaled and delayed $h_{ZOH}$ by a $u(t)$, the result is $$(h_{ZOH}(t+T/2)-h_{ZOH}(t-T/2))*u(t)=\text{tri}(\frac{t}{T})=h_{FOH}(t)$$
The overall result is a first-order hold filter