For some large documents with many Microformat tuples the extensive use of XPath in the DomUtils class cause Microformat extraction to be slow. I've market this as trivial as it's a corner case.
To reproduce the problem the patch addresses, run the Microformat extractor on the folloing url:
I include a patch that improves performance at the cost of code simplicity. I hope someone who is more involved in the project can decide if it's a good idea to use the patch or not, or maybe address this issue in another way.. The patch replaces commonly used XPath queries with DOM tree traversals. Eg. getting all nodes with 'class' attributes. On my machine the time to parse the given document is reduced from around 105 seconds to 14 seconds.