Since this game is a discrete process, I thought it might be interesting to try different values of $N$ (the number of cards in the initial deck) and $x$ (the number of charge counters added) to see what number of charge counters ends the game fastest for each $N$.
It turns out that the equation $x = \lceil \sqrt{N + 1} - 1 \rceil$ (rounding up to the next integer when $N+1$ is not a perfect square) is a good choice of how many charge counters to put on Grindclock. But there are many values of $N$ for which you can put a different number of charge counters on Grindclock and still finish in the same number of turns.
In fact you can have a surprisingly large range of choices of $x$ for some values of $N$ that all end the game just as quickly as $x = \lceil \sqrt{N + 1} - 1 \rceil$.
For example, if $N = 15$, then by adding $\lceil \sqrt{N + 1} - 1 \rceil = 3$ charge counters you can end the game in $7$ turns (including the turn on which the other player cannot draw a card). There is no way to end the game faster, but any integer value of $x$ in the range $2 \le x \le 5$ will end the game in the same number of turns.
In general, the length of the game will be
$$
g(N,x) = x + \left\lceil \frac{N - x + 1}{x + 1} \right\rceil.
$$
The term $1$ occurs in the numerator because according to the original problem statement, if $x + 1$ divides $N - x$ exactly, the second player will draw the last card on turn
$x + \frac{N - x}{x + 1}$, and will only fail to draw a card on the next turn.
It's not hard to build a table giving the length of the game for various values of $N$ and $x$ and to pick out the range of optimal values of $x$ for each $N$.
One pattern that becomes apparent is that if $N = k(k+2) - 1$ for some integer $k$,
then $x = k$ will result in a game of length $2k$, and no other value of $x$
will end the game as quickly. This appears to be the only case in which the
optimal value of $x$ is unique for a given $N$.