The problem is that HTML::Stripper uses HTML::Parser (which accumulates content) in strange ways and the documentation is not very explicit about it.