Ideally one should try to minimize latency by avoiding multiple hops between the array and the ESXi hosts. Other considerations apart the mentioned above would be use of Beacon probing to detect upstream network failures although note that it's not supported with LACP and requires atleast 3 uplinks for correct reporting. Also check the vendor's recommendations on the use of flow control.
There is also a good series on NFS Best practices done by Cormac Hogan which should be a very good read:
http://cormachogan.com/tag/best-practices/
....hth!