When Parallels is not running, the RAM otherwise allocated to virtual machines is available to the host OS.
When Parallels is running but the virtual machine environment is not, the RAM is still avaiable to the host OS.
When you have multiple VMs set up and only one environment is booted and running, the RAM allocated to the other, non-running virtual machines is available to the host OS + the environment that is running.
If this were not so, I’d be in a bit of a jam, having a show-offy slew of operating system environment: FreeBSD 7.1, Windows 3.11, Windows 95, Windows NT Server 4, Windows 2000 Server, Windows XP, Windows Server 2003, Windows Server 2008, Windows 7, Windows 10, Mac OS X 10.5 Server, OS X 10.6, OS X 10.7, OS X 10.8, OS X 10.9, OS X 10.10, and OS X 10.11 … the early builds of Windows don’t need a lot of RAM but the OS X and later builds of Windows sure do. And this is on a laptop with 16 gigs of real RAM to portion out.