A reasonable case could be made for saying that all of the work that the EDAC subsystem currently does, would be better carried out by a purely user-space daemon, however there are a few future developments which can only be done from kernel space:
Taking failing RAM out of use at the page level (in the hope of catching failing pages before uncorrectable errors occur).
Retrying failed PCI transactions.
Implementing RAM "scrubbing" and "background scrubbing" in software.
Implementing interrupt-driven (probably NMI-driven in most cases) error reporting.
More hardware support, including non x86(-64) - where memory controller EDAC interface is not easily accessible from user-space.
Additionally, carrying out the work in kernel space provides a performance advantage for the current polled operation.