There is a very large corpus of email addresses available: http://www.cs.cmu.edu/~enron/.
In processing even a subset of this corpus, I see numerous 'unexpected RuntimeException' errors resulting from tagsoup throwing on truly awful html. It seems to me that being able to do something with this entire stack would make a good '1.0' criteria for tika's email parser.
|Field||Original Value||New Value|
|Assignee||Julien Nioche [ jnioche ]|
|Status||Open [ 1 ]||Resolved [ 5 ]|
|Fix Version/s||1.0 [ 12317967 ]|
|Resolution||Fixed [ 1 ]|