Does a log-transformation of a count value work within a regression setting? The dependent variable in my study is a count variable that is clustered at lower values with a long right tail. If I take a log-transformation of this count variable, the distribution begins to appear normally distributed. My question is that, if one can take a log-transformation of a count value and the subsequent distribution of log(var) is normal, is this an acceptable change to make for regression purposes (where something like OLS can be used instead of a Poisson, negative binomial, etc.)
 A: 
If I take a log-transformation of this count variable, the distribution begins to appear normally distributed.

You're looking at the wrong thing! It's not the marginal distribution that matters, but rather the conditional distribution.
More important than normality is getting the mean-function (conditional expectation of y as a function of x) and secondly the variance function (conditional variance of y as a function of the conditional expectation) close to right. If those aren't close to correct descriptions its generally a waste of time trying to deal with the details of the shape of the conditional distribution.

is this an acceptable change to make for regression purposes (where something like OLS can be used instead of a Poisson, negative binomial, etc.)

One thing to worry about is potential for 0's, since you can't take log 0. Transformation may be more nearly reasonable when the smallest values are not close to 0.
Taking logs may in many cases produce nearly straight relationships, but typically you'll find that with counts the log-transformation is 'too strong' for the variance (often taking you from something where the spread increases as the mean increases to something where the spread decreases as the mean increases).
In addition it is often the case that the conditional distribution looks somewhat left skew, at least if the mean is small (less so when the mean is large).
Sometimes, though, following such a transformation you may find yourself in a situation where the conditional variance is close to constant. If the conditional distribution is not tii strongly non-normal, an ordinary regression might work pretty  well.
One difficulty is inference back to the original scale; for example, exponentiating the fit on the log scale will not get you a conditional mean on the original scale.
