It adds another layer of technical complexity.  And, My group manages ESX but another group manages SAN Storage (Politics).  Booting from SAN has its place though.

DB

Some blades are good to boot from the SAN but look at the cost of SAN disks and a pair of mirrored drives.

I agree.

We don't boot from SAN, because it adds just one more link in the chain.

More to configure, more to troubleshoot, more headaches, more money, very little gain..

Stick to booting from local drives, SWAPPING to local drives, and keeping the 'live data' on the SAN.

The only real advantage that we could see was DR at a hot site where the SAN is being replicated.  Plug new machines in or play around with the WWN and up come the servers (Still need similar HW).  That's the only advantage I could see.

DB

I would say that if the SAN crashes you've lost everything anyway.  From my viewpoint it seems that for the most part people are just being resistant to doing things in a new way.  I know a large bank locally that recently removed all of it's local hard drives from it's ESX servers (over 50 of them) because the only disk related failures they have experienced in the last 12 months have been related to local disk.  The SAN has been the far more fault tolerant than any local disks could ever hope to be.  And with dual fiber paths into the SAN your better off than having an array controller that can go bad.

Just my opinion, but I think you guys should re-evaluate your position on not booting from the SAN.

We tested the multipath ability by pulling out the fiber while things were running with multiple VM's perfoming tasks.  Everything worked flawlessly.  We put the fiber back in and it all moved back.  We pulled the other fiber and again all worked as designed.  We are using an EMC Symetrix and an HP blade center with dual port Qlogic cards.  Perhaps that is why ours works so well.  We haven't tested it on a normal server so I can not say for sure what would happen there.  I just assumed that if it worked on one platform it would work on all.  Can't say for sure.  We will test it when wee get a chance.

Just don't be in too much of a hurry to poo poo the idea on conjecture that it might not work. 

Also as to the cost of hosting the OS on the SAN, we feel that it is a minor expense when compared to the nearly 30 TB of storage we are currently using for data.