I changed the computation of font size for the site-wide tag cloud of researchr to use log2 scale instead of the linear distribution that I used before. This brings out the tags with a respectable number of tags, say more than 50, which were lumped together in the lower ranges. The screenshot on the left uses the old linear distribution, the right one is based on the new log2 distribution. The exact font sizes assigned to the points in the scale may require some more tweaking.