I have done some fsck benchmarking. I recovered an 80% full 250gb volume that was doing a buildworld with 8 parallel processes in .9 seconds. The traditional fsck took 24 minutes. For the vast majority of workloads an unsafe shutdown will not prolong the boot by more than a few seconds. I also shortened the time that entries stayed valid in the journal which allowed me to trim the maximum journal size to 32MB. This will limit the maximum possible recovery time while still providing for 1 million outstanding transactions.
The project to date has taken 370 hours. The patch adds 11,000 lines and removes 2,000. This is much bigger than I was anticipating, although of course lines of code can be misleading.