The existing support to use a specified number of cores for compilation was setting the variable CMAKE_BUILD_PARALLEL_LEVEL but not exporting it, so CMake was not being affected, while the -1 option did work (as it was exporting the value).
