How to beat high load on nfs server

Question

Debian

About the Intel® Xeon(TM) Server
MP CPU 3.16GHz

cat /proc/cpuinfo | grep process | wc -l
eight

free-m
             total used free shared buffers cached
Mem: 28203 27606 596 0 10789 9714
-/+ buffers/cache: 7103 21100
Swap: 24695 0 24695

RAID controller

              *-storage
                   description: RAID bus controller
                   product: MegaRAID
                   vendor: LSI Logic / Symbios Logic
                   physical id: 7
                   bus info: [email protected]:13:07.0
                   logical name: scsi2
                   version: 01
                   width: 32 bits
                   clock: 66MHz
                   capabilities: storage pm bus_master cap_list rom
                   configuration: driver=megaraid latency=32
                   resources: irq:134 memory:d8ff0000-d8ffffff(prefetchable)

Drives:

10x148Gb SCSI U320 15k
/dev/sdb1 807G 674G 93G 88% /storage
/dev/sdb1 /storage ext4 defaults,usrjquota=aquota.user,grpjquota=aquota.group,jqfmt=vfsv0,noatime,nodiratime,noacl,errors=remount-ro 0 1

network cards

ethtool -i eth0
driver:tg3
version: 3.116
firmware-version: 5704-v3.36, ASFIPMIc v2.36
bus info: 0000:10:02.0

ethtool -i eth1
driver:tg3
version: 3.116
firmware-version: 5704-v3.36, ASFIPMIc v2.36
bus info: 0000:10:02.0

ifconfig
bond0 Link encap:Ethernet HWaddr 00:0f:1f:ff:d6:4d
          inet addr:192.168.15.71 Bcast:192.168.15.255 Mask:255.255.255.0
          inet6 addr: fe80::20f:1fff:feff:d64d/64 Scope:Link
          UP BROADCAST RUNNING MASTER MULTICAST MTU:1500 Metric:1
          RX packets:1062818202 errors:0 dropped:3918 overruns:0 frame:0
          TX packets:1041317321 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:10000
          RX bytes:258867684559 (241.0 GiB) TX bytes:396569192650 (369.3 GiB)

only nfs-kernel-server is running on the server

uname-a
Linux nas2-backup 2.6.32-5-amd64 #1 SMP Sun Sep 23 10:07:46 UTC 2012 x86_64 GNU/Linux

Debian 6.

What we have… once a day or two… LA load is growing by leaps
and bounds…
But after restarting nfs LA drops to 1 or 2

after 10 hours it will start to grow
although not always so

— the server is connected by a link to DLINK (not a managed switch)
further to Dell chassis is connected to this switch
on the servers that are installed in this chassis on the dropped:0 interfaces,

but the servers that are connected to the
CISCO Catalyst 2948G (it seems) are dropped: xxxxx The

question is ... why is the LA load growing, which is noticeable on reading and writing ...
on servers that mount /storage

where to dig... where to dig... where to poke... I've already tried everything