I think there are several factors which contribute for increase in the R-value for crystal structure data. If your crystal is twin, then that may have higher R-value. Another thing which may cause higher R-value is disorder in structure, which can be minimized by collecting the same data at lower temperature (77 K, liq. N2 temperature). Does all the atoms in crystal structure are "aniso " and CIF has been cleared with all the errors?
First your crystal quality should be high, which can be indicated by good diffraction spots, large number of unique reflections and small Rint values. (Of course, there are many other factors need to be considered during the data collection process.)
If you have good diffraction data, several factors should be considered during the refinement process. I suppose you know the refinement process. Firstly, you need to consider background, profile fitting function, etc. After this, please consider disorder (many kinds of disorders, such as occupancy disorder...you should deal with different disorders with different ways, you can search for manual on internet, in single crystal case, SHELXL manual) and anisotropic thermal factor (Normally the default set is isotropic, which will give additional electron peak and hole around the atoms which anisotropic osscilations). In addition, you need to consider extinction parameter (EXTI) as well. Sometimes, you can shrink the data range used for analyses (which means small range of theta) in order to reduce R value, but in this case, the number of unique reflections will decrease as well, you need to find a balance.
Look into your sentence and see whether it makes any sense.
You cannot reduce the R to get good crystal??
Good crystal without any defects in it and good data collection and proper structure refinement will automatically reduce the value of R to an acceptable number.
Here's a clever way to get a very low R factor. Measure your sample with a weak source (be sure to remove any background shielding and if the detector has an energy discriminator be sure to open the window as wide as possible). When you have a very low signal to noise ratio you are mostly fitting the background. This will allow you to get a staggeringly low R value even if your model is wrong!