...and here follows some revised profiling data.  Unsurprisingly,
ufs_lookup still dominates completely:

  %   cumulative   self              self     total           =

 time   seconds   seconds    calls  ms/call  ms/call  name    =

 72.91     41.66    41.66   185447     0.22     0.23  ufs_lookup
 15.03     50.25     8.59                             mpidle
  2.66     51.77     1.52    45093     0.03     0.03  newdirrem
  0.89     52.28     0.51    53017     0.01     0.01  scanc
  0.58     52.61     0.33    55383     0.01     0.01  memset
  0.40     52.84     0.23   595496     0.00     0.00  i386_copyin
  0.30     53.01     0.17    37054     0.00     0.01  scsipi_run_queue

Regards,

- H=E5vard