IMO is that such claims are just silly. If the upsampled DSD stream really sounds different than the original PCM stream, it could only be degraded, not improved. The original PCM is and remains the gold standard of that recording session.
Now if the original recording was done in DSD, the claims of improved SQ over a PCM recording done in tandem, might have enough possibility to do bias controlled listening tests and measurements. Personally I don't believe that DSD is superior in any way, but that just my opinion.