CpG content varies a lot across the human genome. What does determine a promoter to be silenced through DNA methylation: the fraction of CpG sites that are methylated or the total number of methylated CpG sites? Any reference?
Generally, robust silencing by DNA methylation requires that all or most CG sites be methylated and that there are a good number of CG sites to methylate. (A 'good number' is relative to each species depending on CG content of the respective genome.)
A promoter with ~10% of its CG sites methylated would not likely be silenced. However, the caveat with this is that one well-placed methylated CG can cause silencing or at least decreased promoter efficiency. Also, many promoters are depleted for CG, which means they can be fully methylated, but still not have much absolute methylation. These promoters can be active despite apparent 'hypermethylation' in the CG context. Also, a caveat to this is that promoters with zero methylation can be silenced, even if they have CG sites. Basically, if CG methylation is important in the regulation of a gene, it will have CG sites in functionally-important parts of the promoter. The promoter will be well-methylated when the gene is silenced, but because placement of de novo CG methylation is not absolutely perfect, there is a high likelihood that there will be more methylation than is needed for silencing to be achieved. A sort of overkill phenotype.
This paper from the Schubeler lab is a nice foray into the mammalian promoters, which they break into 3 different groups depending on CpG content, and then further analyze each group for CG methylation profiles.