How to link processing power of old computers together?

There are a number of schemes for distributed computing. One is called Digipede. At a previous employer (this is in the 90s), we had a warehouse of older desktop computers that hadn't been fully depreciated (even though 75-100Mhz pentiums were worthless by then), and I wasn't allowed to order new servers for the processing my department had to do, so I got a bunch of them, refactored much of the code into DCOM objects. I called it RAIC - redundant array of inexpensive computers. Painting numbers on each of the stack of computers, it was easy to say "meh, go reboot #5." Looking back it was cool, frugal and a waste of time.

My advice would be to not bother. If you have a task that is well suited to distributed processing, you'd already have ideas. If you don't have such a task, you're going around with a hammer looking for nails. In such a case, set them up to run seti@home processing.

if they have enough processing power for browsing and wordprocessing, why not install OpenOffice and put them on Freecycle for someone who can't afford to buy a computer.

Condor is excellent for cycle scavenging.

Using TORQUE/pbs is also very popular for clustered computing. This is a regular package in Debian, Ubuntu and probably many other Linux distributions.

TORQUE, pbs and the excellent scheduler Maui are well documented at Cluster Resources.