I haven’t seen this basic truth described about tag clouds, aka weighted lists, though we have a nice blogdom exploration of the cognitive implications of folksonomies and plenty of contrasts with hierarchies.
Tag clouds are an extremely rational UI design, trading off efficiency of access (font size) against frequency. This optimizes user effort, independent of task. They take Fitt’s law as a literal rule, directly mapping probability of use to UI real estate allocation.
Most clouds are alpha-sorted, but I often use frequency sorted. For individual directed tasks, it provides a non-linear but memorable dimension on which to adjust your scan. And, it’s a fun cognitive experience to travel the dimenson of decreasing popularity/importance. Alphabetic organization is great for where the user has strong knowledge of your tag domain and as the tag set enlarges. I prefer to facet tags to keep the set size small.
Others are focused on dimensions of time in clouds, where interesting representations of velocity may force new variations on tag cloud display.
I’ve had a number of UI traditionalists cast skepticism at the weighted list, but in experience with users, they get it and it’s at a minimum trendy if not actually more fun than the average UI.