A completely different idea to solve my actual problem: Someone else suggested to just take out the conversions all together. I mean, I am converting right back into the encoding I converted from. I have been assured that no link uses a character above the first 128 (7 bit ASCII). As far as I know there are no HTML control characters outside of 7 bit ASCII either. So shouldn't the parser just be able to parse the ISO-8859-1 document as if it was utf-8? Yeah, I know it sounds horrible, but as far as I can tell it should not actually break...
As author of the module: Could this work? What would I have to change in the code to keep any input conversion from happening? (I will play around abit myself, but I am not familiar with the code, nor with Apache module logic. And its been quite a few years since I last coded C...) At the very least this would tell us (if it works) whether or not the conversions are to blame for the problems I experience. Martin