Agreed regarding potential breakage if anyone's only checking the incorrect value, rather than looking for both the known incorrect and documented values. An alternative could be to update the doc with the actual value. Your call!
EDIT: I imagine most people will just check for the presence of the valid value, so IMO it's low risk to just fix this in the data. If anyone's checking the "dead" value, they are probably in the error reporting path.
I have checked the SHM specification and actually it's correct. The values are specified in C-notation as "HWiS" (string) and 'DEAD' (binary value). 'DEAD' in little-endian format translates into "DAED" ASCII string.