"As a matter of fact, the 3 very high level engineers I know all have been buying CRAY lately."
well, first, i said "in the long run", so that point is moot. i'll concede that there's lots of software written for the cray and its easy to program for. however, cost/performance isn't there and with the cheapness of beowulf-like clusters, the motivation to develop huge amounts of public software for them (academia, fx/film industry, gov labs), and the fact that supercomputers built from commodity parts will - history shows - win out over these vector machines ... there's not a huge future there. oh, also see "grid computing".
not sure what your engineer friends do. as of at least 2 yrs or so ago, something like 85% of cray business was government. sgi's big gamble with them never panned out because ... well, because of what i outlined above: their gamble was on high end graphics, but all of the major effects houses dumped sgi/cray machines and went with first sun, and now linux clusters/farms.