Just to be sure, a first step would be to check if you can repeat the design of the primers pairs from http://www.urogene.org/cgi-bin/methprimer/methprimer.cgi, to make sure the ones you have correctly bind to a methylated product.
If the primers are correct, then you could try a gradient PCR to optimise your annealing temperature, and also play with different concentrations of MgCl2 in the reaction.
The addition of Q solution (Qiagen) may also help with difficult to amplify CG areas. I add 5ul of in 25ul PCR reaction tube.
Finally, it could be that you don't have enough DNA to start with? You may lose some with the bisulphite conversion and purification, so adding more cycles to your PCR or making a nested PCR may be useful for you.
90 % GC? Are you sure? Human fmr1 gene ( ENSG00000102081) has about 40 % GC in canonical CDS. Maybe you want to amplify some promoter region? These regions often tend to form very stable noncanonical DNA structures like guanine quadruplexes. Try to increase time of denaturation in PCR cycles and also increasing extension times would be useful. I have some success with robust polymerases for such amplifications (e. g. https://www.sigmaaldrich.com/catalog/product/roche/2grkb).
You are right, I want to amplify promoter region of fmr1 gene in FXS. Basically, fragile X Syndrome (FXS) is a disease in which CGG trinucleotide repeats expand greater than 200 in number (known as full mutation) in 5` UTR region of fmr1 gene, which is otherwise upto 44 repeats in normal condition. This full mutation results in hypermutation at cytosine residues in these CGG repeats and nearby CpG islands of the promoter region.