First, how the data that determines the percentages is open to question and interpretation so they are not in any way shape or form reliable. Second, because the Linux based OS are not, in the main, sold in computer shops like machines with MS Windows and MAC OS/X pre-installed it is notoriously difficult to keep anything approaching a reliable count of installs. The same also applies if they use the web browsers identification string to count install bases and web browser take up as many web sites have broken HTML code implementations so many Linux based OS users fake that string to fool the broken web site into thinking that the browser is running on a Microsoft platform.
I have no real idea of what the real percentage counts should be but I do know there are many second and third, fourth,, fifth etc install bases that are Linux based with only the primary machine being the Microsoft based one because the owners of like to play games that are not available for the Linux based platform. Further, there are many machines that act as a firewall built on the Linux based platform that will not be included when collating such data. Other Linux based installs act as file and print servers, these would not be counted either.
Without falling into advocacy mode it is not hard to work out that there are many more Linux based OS installs than any such data gathering exercise could ever hope to evaluate in a fair and none bias manner.
All in all. Such data should be taken with a large dose of salt and as long as one is aware of why that should be so then we can be assured such data is largely ignored for the fodder that it really is.