Hi,
1) When is a good time to start building the peptide into the density (should I wait until Rfactors improve significantly)?
I would advise waiting until the rest of the model is as good as possible. That way you can use the difference map as a paper figure without needing to do any extra refinement to remove model bias later.
It's good to keep in mind that in this case map will be biased by the non-atomic model (bulk-solvent), as well as by the atomic model as shown here: Acta Cryst. (2002). A58, 270-282. Also note, more (correct) model you build - better map you get - so even more model you can build.
3) What is a general good strategy for lowering Rfactors? (Esp if the gap between Rwork and Rfree is high..like 28/40 or so)?
28/40 sounds worrying. In majority of scenarios I've seen so far such a big difference between Rfree and Rwork indicates a problem that is typically out of scope of refinement. Pavel