It depends to which kind of representativity you are looking for. There is something like 600.000 projects taken into account by Ohioh, so you can consider it as a big enough data set. However, some of these projects are not relevant, and they are often more numberous than believed. For instance I know in Gnome (which seems to be indexed), about 60% of all the projects didn't receive any commits during the last 2-3 years. You have to throw them away if you are interested in "modern" trends.
An other point is that the number of projects is maybe a too coarse metric of the language popularity: Most of the projects are very small, and a few are very big (their sizes typically follow a Pareto distribution). So in fact you will probably be more interested in the cumulative size of the relevant projects or the projects activity. You have then to define what's a project size (number of lines of code? number of files?). If you look at the monthly evolution of the number of commits per language, you can see that C, HTML and Java are quiete close. And the C activity is slowing down since 2006. If you look the number of projects that received at least one commit per month, the C language is even in the second place, after the HTML language! The monthly number of lines of code changed is probably affected by important refactoring/restructuration in some big projects.
I fear the data are also very heterogeneous. Eclipse and Apache are probably pro-Java, since most of KDE projects will probably written in C/C++. The Linux kernel is mainly written in C. So if you focus on a particular subcommunity, the results may sharply change.
At the end of the day, the presented data can be relevant for a trend analysis, but you have to clearly define what you are looking for and adapt your observations consequently.