I think the idea is that you take a list of all domains and then count the number of DNS lookups that are done for each over the course of some time period (e.g. 1 year). Then sort them from high to low number of lookups. At the start of the list you'll probably find only a few domains with many billions of lookups. As you go down the list you'll find many domains with very few lookups, the "long tail".

It's a bit confusing because normally "long tail" refers to a histogram or probability distribution where a large portion of the population is far from the central part. I don't think that works in this case, unless I'm confused about what to put on the x-axis. (Because if the x-axis is lookup frequency, the domains being referred to here would be in a peak close to 0, not in the tail.)

edit: maybe the x-axis of a histogram could be "mean time between lookups" instead of frequency? That would put the popular domains in the peak near 0, and the unpopular ones farther out in the "tail".