Yep, it has been scientifically proven: the accuracy of people in describing charts with 'chart junk' is no worse than for plain charts, and the recall after a 2-3 week gap was actually significantly better. In addition, people overwhelmingly preferred 'chart junk' diagrams for reading and remembering over plain charts. In all, the researchers conclude that if memorability is important, elaborate visual imagery has the potential to help fix a chart in a viewer's memory.
Update: On Statistical Modeling and Causal Inference, Andrew Gelman draws attention to a couple of serious flaws with the argument: the plain graphs compared to the chartjunk graphs just aren't very good; chartjunk severely limits the amount of information that can be included in a graph. The whole thing here.