Parse is 1.1-1.2x faster -- zerodep now edges ahead of xmltodict across all document sizes, including small inputs.
Unparse is 1.4-1.6x faster -- serialization is where zerodep shows a clear advantage, with the gap widening on larger documents.
extract_tags has no competitor -- this is a unique feature for LLM output parsing. The first_only=True optimization is extremely fast (~7.7 μs).
Both use expat -- unlike YAML/JSON where zerodep reimplements the parser, both zerodep and xmltodict use the same C-based expat parser underneath, so unparse speedup comes from more efficient string construction.