But I think the error needs to be much larger than you're likely to get from processing the data for the refinement to suffer significantly. It definitely doesn't explain the discrepancy in R-factors, since even an excellent protein crystal diffracting to atomic resolution - where there is no doubt as to the unit cell parameters - will still have this problem.
Yes, it does seem that the size of the R difference might be small, but one could also imagine a beta-strand, for example, aligned along one of the cell axes, and "being told" one thing by geometric constraints, and another thing by a cell constant that was off by, say, 0.5%. If the putatively-imprecise cell constant information were removed, the strand might "relax," and the R value might drop a bit. It seems to me that cell parameter measurements are always fairly imprecise due to having a lot of experimentally-variable parameters. And being able to predict spots on the detector is no proof that the parameters are exactly right, I don't think. Jacob