The reason the market for this is weird is the same reason Sun took the DES chips off its motherboards years ago -- you can't conduct modern business with the fucked up export regime we are dealing with.
Also, even if national boundaries could be transcended (it's not just the US any more on this score), I'll grant to Tim that server-oriented cryptography h/w isn't going to be a mass-market item. It's going to be a niche market, but it's an _absolutely vital_ niche market if this stuff is going to take off in a big way. I, too, wouldn't invest big bucks in a company that did nothing but server-oriented crypto h/w, but I know that a lot of very good business ideas _won't work at all_ if this stuff doesn't exist. At the same time, there is going to be a very good business in consumer-oriented crypto devices as the problems inherent in using a general purpose computer for storing & processing keys, e-cash, etc. become apparent. I'm leaning away from the less specialized stuff, e.g. "checkbook on a PCMCIA card" and toward general purpose cards like the nat. semi. and telequip stuff. Both of these companies are rumored to be offering server-oriented products in the next couple of quarters to compliment their consumer product lines. The problem with general purpose machines has nothing to do with native processing vs. DSPs, but rather the fact that g.p. machines have mechanical hard drives and don't fit in your pocket. They also tend to house a lot of dubious other software that could get its grubby hands on things. These issues of reliability, portability and security don't affect things like soundcards, which is why native signal processing is likely to win there, but not in the use of crypto to secure valuable transactions.