At Google we noticed a change in behavior for the HKSCS charset family between jdk7 and jdk8.  It looks like a bug in the jdk8 implementations.

Recipe:
-----
public class HKSCS {
    public static void main(String[] args) throws Throwable {
        StringBuilder sb = new StringBuilder();
        sb.append((char) (Character.MIN_HIGH_SURROGATE + 67));
        sb.append((char) (Character.MIN_LOW_SURROGATE + 67));
        sb.append('a');
        byte[] xs = sb.toString().getBytes("Big5-HKSCS");
        for (byte x : xs) {
            System.out.printf("%02x ", x & 0xff);
        }
        System.out.println();
    }
}
---
Produces:
8a a6 61 
in jdk7 and
8a a6 
in jdk8, which seems wrong.