On a UFS2 filesystem with soft updates, deleting a directory of thousands of symlinks is hundreds of times slower than deleting regular files.

It looks like the writes are synchronous.

  $ mkdir test1 && for f in `seq 5000`; do touch "test1/$f"; done
  $ mkdir test2 && for f in `seq 5000`; do echo test > "test2/$f"; done
  $ mkdir test3 && for f in `seq 5000`; do ln -s test "test3/$f"; done
  $ sync
  $ time rm -r test1
          0.20 real         0.01 user         0.17 sys
  $ time rm -r test2
          0.30 real         0.00 user         0.25 sys
  $ time rm -r test3
         94.73 real         0.02 user         0.72 sys

But if the symlinks are made large enough that their targets cannot be stored in the inode directly, then unlinking them is fast!

  $ test="$(perl -e 'print "x"x1023')"
  $ mkdir test4 && for f in `seq 5000`; do ln -s "$test" "test4/$f"; done
  $ time rm -r test4
          0.17 real         0.00 user         0.17 sys