Because NFS is not SCSI based locking like in FC there will be no single I/O contention point (something we stuggle with on the current MSA we have! 1.5TB LUNS 40 VMS a LUN!!), in theory it should not make any difference to the Filer if its dealing with 60 VM's in 1 volume or 60 VM's across 3 volumes. Data I/O rates will still be the same. I think this is compounded by the NetApp documentation that sates no limit to max number of VM's inside an NFS volume.
I think the only requirement to make volumes smaller on NFS is from a Snap/replication point of view. Our infrastructure will require 98% of our VM's to be replicated to our DR site so seperating VM's in volumes isn't an issue. Anything that doesn't need repl is generally a test box and I will probably store on SATA anyhow. Also, the default value for max NFS Datastores is only 8 and the option to only increase up to 32, kind of implies that lots of little ones are not needed.

To put this into context in our environment (A 2 head 3140, 2 shelves of FC disks, 1 SATA shelf);

Using a 14 disk shelf as an aggregate with 1 hot spare and dp, that equates to 2.5TB formatted storage. Our NetApp guy said you dont want to push an aggregate much beyond 85% full, so thats now 2.1TB of useable NFS space.
I'm working on an average C: size of aprx 20gb, swap size of 2gb, and currently we have 1.2TB of data drives allocated to vmdk's but only 500GB used inside the vm's (ignoring our file servers which are going CIFS on SATA).
I'm currently thinking along the lines of: 2.1TB (aggr0) - 1.5TB (esx_system01) - 150gb (esx_swap01) - 500gb (esx_data01) = as good as "0" so thats an 85% utilized aggregate for arguments sake.
With 20% Snap reserve on the system and data volumes (0% on swap) we will have 1.2TB for C:'s (so about 60 VM C:) and 400GB for data.
I think thats probably a good compromise. If I do that on each shelf we will have 6 NFS volumes to present to our 14 ESX hosts, and the potential storage for roughly 120 machines (we have 80 currently).

I think for me the question of huge volumes or not will arise when our environment expands. Having only 2 shelves and 1 active on each controller means I can't make really huge volumes right now. But, if I add another active shelf to each controller the option to double the aggr to 28 disks will be something to consider. I think the extra spindle speed may well be welcomed???

Be interesting to hear others thoughts and even setups with NFS. Just need to sus the correct setup now for seting up the ESX hosts for aggregated links so we can use 2GB.

Cheers,
Rich.

Edit: I had forgotten about ASIS. Later on we may dedupe the whole VI cluster, but I'm currently not sure on how this effects running performance of a VM. So for now I have on one of the shelves only, scaled back the (C: esx_system01) volume to 800GB and created another (esx_asis01) 400GB volume which will fit our 15-20 lower use servers. Guess the beauty of NFS is being able to shrink and grow so easily if the sizes end up being way out.