Yesterday's UTF-8 fix was pretty much the textbook example of yak shaving. It turns out there's a bug in the version of libxml2 in the standard heroku cedar stack (2.7.6) that clobbers UTF-8 characters when manipulating documents.
The solution ended up being to upgrade to the newly-released cedar-14 stack, which is based on a newer Ubuntu LTS (14.04) and has a number of upgraded packages. Crazy to think that the standard Cedar stack is based on Ubuntu 10.04, which was released in April 2010. Doesn't seem like it's been that long. Time flies.