For example, try actually making sense out of the second diagram on that site. It's far harder to read than the first. You can actually only make sense of it by cross-referencing with the first diagram.
What kind of logic is that?
The question IMO is not whether it is visually misleading but whether it conveys information in a usable form, and whether the information it conveys in that form is misleading, and I don't see it.
If we adopt a visually perfect line doesn't that mean all graphs must be linear (no plotting on non-linear scales, even though this is useful in some contexts), start with an origin of (0,0) and so forth? What percentage of graphs do you see that obey these rules?
The Nielsen graph clearly represents 51% Android as about 33%, apple's 34% as one quarter and the Blackberry als one-fifth. If you prefer readability of the labels over more-or-less-correctness of a graph which suggests it covers a full 100% and it's subdivisions you should maybe just use a table. As in Excel. A 30-31 x-scale is misleadin but showing 10% as 20% though is more than sloppy because you do show the "origin". This has got nothing to do with visually perfect.
While making for bad UX, it doesn't change the truth of the information at all, and it would be easy to fix these issues.
Edit: Looks like someone already has: http://9to5mac.com/2012/07/13/nielsen-needs-to-work-on-their...