Greg Restall has started a Curry-Howard wiki called "Proof and Counterexample"; it seems to be some kind of consultation exercuse for his planned book. I am enormously happy to hear this, which I heard of via, where else?, Lambda the Ultimate.
An Answer, sort of
Thomas Ptacek pointed me in the direction of ent, which isn't quite the low resource estimator for gzip performanceI was after, but it's in the right sort of direction.
Fallibility
I've been making lots of assertions that turned out to be false in my last few diary entries. I'm feeling particularly raw about misidentifying the SYNTAX-CASE records implementation...