No, this isn't normal, even with BS-seq datasets where base qualities tend to be a bit lower. I've just looked at a couple of mine and they don't show anything like that. Having said that, I wouldn't worry that much about this. It's likely that your alignment efficiency will be a bit lower, but this is probably still a salvageable dataset. When you get to the methylation extraction step of your pipeline, be sure to use an extractor that can filter according to base phred quality. If whatever you normally use doesn't support this, then just use PileOMeth.

Edit: BTW, are these Illumina reads? I assume so but it's good to double check. I expect ion torrent among others has a different error profile and might have issues with datasets like this.