ECC without reporting puts you in a weird place where the observable error rate is less, but any observable error will be a multi-bit error. With reporting, you would likely replace a dimm that regularly has single bit errors, and your system would typically halt on any two-bit errors, and three or more bit errors may or may not be detected. Advanced operating systems may let you kill only the processes affected by the two-bit errors, but afaik that's only for mainframes and maybe commercial unix (Solaris? AIX?).