I understand the question like that:

If you apply an value to an MSR on multiple cores, do you perform the masking for each core or do you use the result from the first core for the other cores?

BTW I like the result a lot better than my proposal. I guess coremask might also work on multi socket machines.