In one of the structures I have refined, I found phenix 1.5_2 gave 1 to 1.5% lower Rfree and Rwork than  REFMAC 5.5.0102.  
Can I propose in a publication that  "We attribute these differences to methods for handling bulk solvent in PHENIX that are less adversely affected by deficiencies in the crystallographic data."? 
This is a guess, because there is a large contribution to refinement from bulk solvent, rather than factual information. I did cross-refinement, and found it going from structures refined in REfmac to Phenix, and vice versa. The Phenix maps looked better, too. 
Is this a reasonable suggestion, and would there be a relevant reference?
Tim