It's the old Windows regex issue (http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6337993) that raises its head once more, this time in the examples tables. It seems to be related to the number of rows and how similar they are amongst each other. It also seems to be sensitive to the content of the first couple of columns, i.e. if the first and/or second column content differ the problem does not always manifest itself.
Solutions: either refactor regex parsing to not use patterns such as (x|y)* or provide an alternative implementation or parser based on a pluggable regex impl, different from the JDK one.