I thought there was a post asking the question in the title, but cannot find it anymore.
It is not difficult to construct a process with countable many continuous sample paths, all starting from 0, but has discontinous variance at 0.
So, I will add some additional conditions:
Let $X_t$ be a Markov process with continuous sample paths, prove (or give a counter example) that $Var(X_t)$ is continuous with respect to $t$
Also, how about the case where $X_t$ is a martingale instead of Markovian? Perhaps the time-changed Brownian motion theorem can be helpful?
[Edit] come to think of it, in the Markovian case, even the continuity of $E(X_t)$ is not obvious. Any known results?