I'm using it to debug a problem.

I recommend Devel::Peek's Dump.

The corruption is consistent with bytes with the high-bit set getting converted to their UTF8 encoding.

Way too many XS modules access a string's buffer with no regard to the setting of the UTF8 flag. Sounds like it's happening here.

string	possible internal representations	default typemap to char*
"\x{C9}"	C9; UTF8=0	C9
	C3,89; UTF8=1	C3,89

I suspect using utf::downgrade is the solution, but I want to verify that.

It is indeed the solution. utf8::downgrade converts the internal encoding of a string from utf8 to bytes if it isn't already. The XS module should do that for you, but you can do it for the module.

string	possible internal representations	downgraded	default typemap to char*
"\x{C9}"	C9; UTF8=0	C9; UTF8=0	C9
	C3,89; UTF8=1	C9; UTF8=0	C9