It's been several years, but in my experiments it felt plenty fast if I prefetched links at page load time so that they're already local by the time the user actually tries to follow them (sometimes I'd do this out to two hops).

I think it "failed" because people expected it to be a replacement transport layer for the existing web, minus all of the problems the existing web had, and what they got was a radically different kind of web that would have to be built more or less from scratch.

I always figured it was a matter of the existing web getting bad enough, and then we'd see adoption improve. Maybe that time is near.