It might seem like that's something you would want, but you would be very unhappy with the result. The network connection between machines is many thousands of times slower than the communications between processors and memory in a single machine. Fooling the OS (with a hypervisor) would be especially bad.
That's why cluster application must be aware of the cluster, or use multiple processes (that can be migrated to other nodes in the cluster) that use network communication and disk files for data exchange. Without that awareness/architecture, the performance of the applications would be significantly lower negating the benefit of clustering.
The type of cluster you are thinking about is probably a High-performance (HPC) cluster. There's more information at Wikipedia