I agree in principle. I don’t know exactly where the “best” value of k is, but I’m sure it depends on the actual ranking systems used and their variability.
In the end, the goal of choosing a high k is to make the difference in value between a tank of 1 and a rank of 2 to be close to the same as the difference in value between a rank of 1001 and a rank of 1002.
There are, of course, other schemes to reduce these differences. For instance, rather than doing the harmonic mean, you could do the generalized mean
M_p(x_1, x_2, \ldots, x_n) = \left( \frac{1}{n} \sum_{i=1}^{n} x_i^p \right)^{1/p}
with p = -\frac{1}{2},
M_{-\frac{1}{2}}(x_1, x_2, \ldots, x_n) = \left( \frac{1}{n} \sum_{i=1}^{n} x_i^{-\frac{1}{2}} \right)^{-2}
Alternately, if you wanted to do the opposite and increase the weight of high-ranking results, you could use p = 2,
M_{-2}(x_1, x_2, \ldots, x_n) = \left( \frac{1}{n} \sum_{i=1}^{n} x_i^{-2} \right)^{-1/2}