About two years ago, I made a post-processing filter for Data::Dumper which would make the output more legible. Not as radical as going breadth-first instead of depth-first, but it helped format the output much more readably for human consumption.

As with you, I found the C-and-Perl issue to be daunting. If not in skills required, then in hassle to overcome. So I just decided to not bother. I would have to rework the main loop of Dumper to facilitate the filtering, and I just couldn't be arsed to jump through that hoop.

It's also part of the main distribution, now, so any changes have to be checked just that much more deeply. It's basically frozen in time and won't get much better than it is. Which is a shame.