In designing the HP(E) Superdome, where possible, we tried to do end to end protection. There are some areas where that is not possible (like when you use Intel processors, unless you use voting). Double chip-kill did a lot to improve reliability by allowing failing dies to be bypassed. Superdome used double chip kill because many of its targeted markets were running OLTP systems where corruption was considered unacceptable. Removing failing DIMMs caused too much down time. But if you take a workload off of a Superdome and spread it across multiple small servers you may open yourself up to many more SDC events, even if you have ECC. Memory SDC likelihood is based on the amount of memory used and the altitude (and other factors).

BTW I had heard that some Intel processors chip kill could break if there more than 4 memory buses used because of resource constraints in the correction mechanisms on the on die memory controllers. I hope that was fixed or never existed.