Through simulation, I've noticed that estimates of the coefficient of variation (CV) of exponentially distributed variables are biased at low sample sizes (as seen in the plot I made). I've seen an equation for calculating an unbiased CV for small samples which are normally distributed. Is there something similar for exponentially distributed values? All my google searches are coming up dry.
Check out "On New Moment Estimation of Parameters of the Gamma Distribution using its Characterization," Ann. Inst. Statist. Math., 2002, by Tea-Yuan Hwang and Ping-Huang Huang, available online.
From their work, I think you should try this estimator, where $T$ is the CV, $n$ is the sample size, $S_n$ is the sample standard deviation, and $\bar x$ is the sample mean: $$\hat {T}=\sqrt{\left( {{n+1} \over {n}} \right) \left( {S_n^2 \over \bar x^2} \right) }$$
$\hat T^2$ should be unbiased for the square of the CV. Not quite what you want, but an improvement over what you currently are using.
They mention some exact results in another reference, but only for sample sizes of 3, 4, or 5.