Corrupted non-ASCII charachters in a JaCoCo report

It seems like the JaCoCo plugin now uses a platform default enconding (Cp1251 in my case) to load sources and pack them into HTML (build/reports/jacoco/test/html/**/*.java.html).
I have a project with sources that contain non-ASCII charachters (encoded in UTF-8) and these charachters become corrupted in the output HTML.