Internal conversion between the server character sets preserves the error character.
For example, when translating from LATIN to UNICODE to KANJISJIS, 0x1A translates to U+FFFD which translates to 0x1A.
For the KANJI1 character set, conversion of multibyte characters never causes an error.