Does that leave an open question then as to whether or not the overall responsibility for XSLT 2 conformance relies with the XSLT 2 processor even though the bug is attributable to the XML parser?

Yes, it's an open question, but it's an engineering/contractual question,
not really a spec conformance question.

In practical terms, writing an XSLT processor that's resilient to arbitrary
bugs in the XML parser is not a practical proposition. And the bugs in the
Sun version of Xerces are pretty arbitrary...

I certainly agree with you that, practically speaking, it would be a waste
of MIPS for the XSLT processor to have to baby-sit a buggy XML parser.  And,
yes, particularly with open source that users don't have to pay for, the
integration responsibility should rest with them.

Cheers
Justin Johansson