This is the remaining failure in resque's 1-x-stable branch in 1.9 mode. The following code should work in 1.9 mode:

str = "\xFA\x8D\xA8\x8A\xBF4\xD3\x9FV]_"
str.force_encoding('BINARY').encode('UTF-8', :invalid => :replace, :undef => :replace)

We have a hard check for ASCII to multibyte that always raises error for 8-bit characters. In the case that :undef => :replace is specified, we should be allowing the encoding process to proceed to replacing those bytes, rather than hard erroring.

Fix pending.