ZFS: kmem_map too small panic again

[Bakul Shah]

The system first paniced during a restore operation to zfs.
Now I get this kmem_map too small panic by doing a couple of
du in the partially restored dir.  This bug seems to have
come back as of yesterday (about the time freebsd namecache
started to be used?) -- prior to that I restored many more
files and did several make buildworlds on the same filesystem
with no problems.

Sources were cvsupped about two hours back (as of approx
10:30pm PDT Apr 25).  I did a sysctl -a on a freshly booted
machine and after one du and diffed them.  The most glaring
diffs seem to be the one shown below (I can supply both
sysctl outputs).  vmstat shows the solaris pool is using
about 127MB.

-debug.numcachehv: 267
-debug.numcache: 2166
+debug.numcachehv: 14517
+debug.numcache: 86100

-kstat.zfs.misc.arcstats.deleted: 16
-kstat.zfs.misc.arcstats.recycle_miss: 0
-kstat.zfs.misc.arcstats.mutex_miss: 0
-kstat.zfs.misc.arcstats.evict_skip: 0
-kstat.zfs.misc.arcstats.hash_elements: 92
-kstat.zfs.misc.arcstats.hash_elements_max: 94
-kstat.zfs.misc.arcstats.hash_collisions: 1
-kstat.zfs.misc.arcstats.hash_chains: 0
-kstat.zfs.misc.arcstats.hash_chain_max: 1
-kstat.zfs.misc.arcstats.p: 83886080
-kstat.zfs.misc.arcstats.c: 167772160
+kstat.zfs.misc.arcstats.deleted: 50263
+kstat.zfs.misc.arcstats.recycle_miss: 7242
+kstat.zfs.misc.arcstats.mutex_miss: 6701
+kstat.zfs.misc.arcstats.evict_skip: 9294733
+kstat.zfs.misc.arcstats.hash_elements: 3514
+kstat.zfs.misc.arcstats.hash_elements_max: 18588
+kstat.zfs.misc.arcstats.hash_collisions: 9805
+kstat.zfs.misc.arcstats.hash_chains: 160
+kstat.zfs.misc.arcstats.hash_chain_max: 4
+kstat.zfs.misc.arcstats.p: 15810418
+kstat.zfs.misc.arcstats.c: 16777216

-kstat.zfs.misc.arcstats.size: 963072
+kstat.zfs.misc.arcstats.size: 57576448