All the current Mac lineup already are supercomputers. A supercomputer is a computer that has multiple processors working in parallel to speed computational jobs up by dividing them across multiple processors rather than a single one. When I run handbrake on my iMac, it divides the job across all 4 cores in parallel, ie it works as a supercomputer.
An extension of the supercomputer is the cluster, in which a number of physically separate computers are connected together to work in parallel. The distinction here is that in a regular supercomputer, all processors are on the same shared bus as all the memory, so that all processors can access all the memory at the speed and latency of the bus. In a cluster, each node has its own independent memory, and the connections between nodes is very much slower than the bus speed. To take advantage of a cluster, not only does the processing load need to be divided into a number of parallel streams, the memory storage needs to be divided too, so that the data passed between nodes is kept to a minimum. This is only really used where seriously large computational power is needed, for example in university research and heavy duty industrial R&D.
While you could build a cluster out of macs with ethernet interconnects, this is a very cost-ineffective way of doing it as the hardware comes with plenty of expensive kit (like screens) that is of no use to this kind of number crunching. A cluster of Mac Pros might be worthwhile (or conceivably minis), but there are far better solutions to the problem out there (with hardware chosen and assembled with this purpose specifically in mind). If you want really big clusters, then you are looking at specialist rack mount kit (think Xserves) and a dedicated staff to maintain the thing.