Right - because we currently limit RAM for our projects to 1GB, compiling some packages will fail. However, we do make an effort to compile, cache and serve up packages from CRAN in a binary format so that you don't need to compile them. When we encounter a (CRAN package & version & R version) we do not have cached, we serve the source package and queue an attempt to compile it.
If the compilation is successful, then the next time it is requested, we will serve a binary package.
If it's essential you pull a version directly from GitHub, then our caching mechanism won't help you, but if you can use the released versions of packages (e.g. tidyr, dplyr etc.), you should be all set.