I've used it to drive a bunch of independent makefiles, when I didn't need an overarching makefile. I've used it to sort dependencies in a simple work queue thing I wrote in the shell. It's not generally all that useful, but being able to grab it and do a quick tsort on paired strings is significantly less painful than writing a Python or Scheme script to do it, and it fits neatly into a pipeline.

I wrote a simple (http://github.com/silentbicycle/wn) to-do graph analyzer for exactly that. I got sick of analyzing "I have to do X, but X needs Y and Z, and Z needs A, and ..." dependency charts. I'm comfortable with BSD make, but sometimes clients give hairy dependency graphs.