According to Sun's Java docs, String.getBytes() will return bytes in the platform's native encoding, which may not be UTF-8. Cross platform (or same platforms configured differently) may end up with problems in the same cluster.

The fix seems to be replace all String.getBytes() uses with String.getBytes(Charset.forName("UTF-8")) to always force UTF-8 conversion.

This bug may also apply to other unrelated places in Infinispan, I haven't checked.