perhaps some researchers are afraid of possible aplification bias? (due to the difference in CG % of the methylated and unmethylated templates, it could be that the unmethylated could be preferentially amplified)
if this is the case, this would even be enhance by your nested-PCR approach; why not a direct one instead?
problems of amplification bias you ofcourse won't encounter when cloning, but with pyrosequencing you'd still see them I think
also, there is the issue of the inherent patterns of sanger sequencing which make quantifying DNA methylation percentages more difficult (I refer you to a question I just answered where I also brought this point up)
Every method can be reliable or unreliable. It depends on how every single step of the protocol is performed (quality sample, dealing with bias,...) and what information you want to extract from your experiments (CpG specific, haplotype specific, overall change,...). For more info see: Article An optimized strategy for cloning-based locus-specific bisul...