Seems to be a problem with incompatible NFS block size on the SMP (NFS client) compared to NFS server. Change your block size on the server to 8K. The default on most Linux servers is 16K.
I think it’s more to do with the switch buffer capacity. If the NFS block size is larger than switch port buffers, it causes a lot of retransmission and slowdowns. I have my block size set to 64K and it still works, but my switches have pretty fat buffers.
one way to test this is to force your NFS server to 100 meg connection instead of gig. In that way, the switch buffers aren’t as heavily taxed.
Looks like you may have a dedicated switch for the SMP. I am using one of the switches/ports on the router and increasing the buffer size will cause latency issues