The Cray-3 was Seymour Cray's big gamble and, ultimately, a very expensive flop.
The design of the Cray-3 is very different when compared to today's personal computers and supercomputers. The components were arranged in all three dimensions to reduce wire lengths (and therefore latency), and the chips were made entirely of gallium arsenide instead of today's silicon transistors. At the time, silicon wasn't advancing fast enough for Cray's designs, and the (toxic) gallium arsenide components could run faster and use less energy.
Like the Cray-2 before it, the Cray-3 was cooled by immersion cooling-- fluorinert filled the whole computer; it was so dense and so hot (over 80kW for a 4-processor configuration) that air wouldn't be efficient enough to stop the machine from melting into a toxic ichor. The fluorinert was pumped at a rate of 10 inches per second, fast enough that erosion became a real concern.
The machine as a whole could peak up to 15 GFLOPS with 16 processors running at 474MHz each, which is about as fast as a modern computer with a dual core processor. It was a failure in the market, with only one buggy computer actually shipped. It was extraordinarily expensive, but the real reason it failed is that the cold war was ending at the time it was released, so the demand for heavy duty supercomputers dried up. That, and the death of Seymour Cray, was the reason there was never a Cray-4.
Fact-checking: http://ed-thelen.org/comp-hist/, old newsgroups