If a database VM needs to compete for the same physical memory space with other VMs, the hypervisor could start allocating memory on disk.

Using vmmemctl (balloon driver), if the hypervisor decides that it wants to reduce the amount of physical memory used by the VM, the driver starts to perform memory allocations inside the VM pushing out other memory data to disk (to the swap space).

This behaviour is highly desirable in general, because otherwise the hypervisor would have to allocate the VM memory data in it's own swap file. In that case the OS believes that it has allocated physical memory, and the application will experience some unexplainable delays and result in performance problems.