This is a common question and there are many different answers. It largely depends on how robust the data is to begin with. Using high sequence coverage, the methylation peaks identified are highly reliable. But to determine a region as differentially methylated between one sample and another, it depends on the degree of methylation- we detect differences that are very clear (and reproducible) since we are comparing a DNA methyltransferase deficient with a proficient cell type. MeDIP-seq by nature is qualitative and should be complemented with quantitative methods (which we have also performed and described). You are correct that MeDIP-seq would be more costly given the sample size needed to detect significantly differentially methylated sites than targeted methods, but our point was that it is less costly when sequenced on the semi-conductor platform than on optical-based sequencing. The sample size will vary depending on the specific question and loci interrogated. I cannot advise on any minimum sample size since it will largely depend on the question(s) of the study and analyses performed. 

i would like to thank you for replying to my query and giving some optional ideas to approach. But the confusion lies the same that during differential methylation between two samples and hypothetically if differences found to be explicit. Can we use that data with few samples for further analysis on all the samples but only on specific genes. 

Yes, that is what I would recommend. I believe it is acceptable practice to evaluate a few case vs control samples for genome-wide differential methylation and then validate local differences using gene-specific techniques; then you can also determine how frequent the differences are in a larger sample size at those loci without enduring the genome-wide costs. 3 case vs 3 controls for genome-wide analyses would be a good place to start, then you can increase if needed or use this data set, if the differences are robust, to start validation at specific loci.