The use of RRBS instead of WGBS is mostly about coverage vs. cost.
To confidently call methylation status at a cytosine residue, you need to have at least 10x sequencing coverage at that particular residue (ENCODE standards require at least 10x coverage for RRBS and 30x coverage for WGBS). To perform whole-genome bisulfite sequencing (WGBS) on large genomes (e.x. mammalian genomes) is very expensive if you are trying to achieve the 10x coverage requirement.
The principle behind RRBS is to improve the sequencing depth of cytosine-rich areas by enriching your sequencing library with CpG-dense regions of the genome. This is accomplished through the use of CpG-specific restriction enzymes like MspI, which recognizes 5'-CCGG-3'.
In RRBS, the entire genome is cleaved using MspI, leaving at least two cytosines per cleaved DNA fragment (one cytosine at either end of the MspI-MspI fragment). Sequencing libraries are constructed from size-selected MspI DNA fragments, bisulfite conversion is performed, and the converted library is sequenced. Specialized software is needed for alignment and analysis of the converted genome (e.x. Bismark, methylKit).
One of the major limitations of RRBS is that you will sequence only CpG-rich regions of the genome. Also, since the library is size-selected after MspI digest but before adapter ligation, you will never sequence any MspI fragments outside of your size-selection range (~40-250 bp).
I have analyzed both types of data and found that the RRBS data was very unsatisfying and didn't answer our research question. WGBS, though extremely costly, is a better option for a comprehensive cytosine modification profiling. There are publicly available RRBS and WGBS datasets at https://www.encodeproject.org/.
I must mention a critical limitation of both techniques in that neither RRBS nor WGBS can distinguish between the type of modification (methylation, hydroxymethylation, carboxylation, etc) present at a particular cytosine residue. RRBS and WGBS, which rely upon the ability of certain modifications to protect cytosines from deamination, can only be used to determine whether cytosine residues are modified or not.
I would suggest following this link for an exceptional technical and practical article in which the authors compared RRBS to WGBS: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC4026711/
A more technical guide to RRBS: http://www.bioinformatics.babraham.ac.uk/projects/bismark/RRBS_Guide.pdf