I find the experimental setup disappointing. As set up, each identifier is surrounded by a fluffy cloud outline, in complete isolation from any surrounding text, testing only the ability to distinguish between identifiers that differ in one or two letters only in the middle (versus the beginning or end, as would more frequently be the case). In practice, long identifiers with such subtle differences are a problem in any naming scheme.<p>What they did not test is the speed of visual chunking of lexical elements, something that is (in my guess) more important than the ability to distinguish between mostly-identical strings. Underscores have fallen out of favor because of this more important visual confusion, where identifier beginnings and endings are less apparent.<p>Of course, this study suggests that our IDEs should perhaps surround variables with fluffy cloud outlines to aid with underscore variables. :)