I have seen Big Blue's booths at broadcast shows selling array processors for compression, but there always semed an immense gulf in intellect between what they were offering (amazing performance) and what could be considered by a regular engineer in a regular company or even start-up (ie a simple starter pack or ref design).
IBM need to encourage some of their own, more restless guys to leave(!) and start up on their own to apply some of these projects (can't see that happening, traditionally, not at IBM).
But with the blessing of the parent company the entrepreneurs might do OK. The real gurus would stay behind in the labs, offer support while working on the next project.
The feedback from the start-ups might focus the parent management and lab guys a little, too.
At last IBM has started providing it in a commercial product, but still as written in the article it required a very efficient programmer to exploit the benefits of the transactional memory provided in terms of hardware.