there is no "gold-standard" for this. Each has it's own underlying assumptions. Calculating by %methylation using the unmethylated control assumes that the unmethylated control is truly unmethylated. Further, that the genomic complement of your controls are equivalent to the samples you are testing (eg: they are genomically balanced), if you have CNV in your samples that deviate from your controls there is potential of mistaking DNA methylation with copy number aberration.
Going with %input is more prevalent because it inherently takes this genomic balance/imbalance into account. However, you are then making the assumption that any enrichment you are seeing over and above your input is in fact methylation and that your antibody is actually binding to, and detecting methylated DNA.
Plenty of literature around both issues and concepts.