Here's how it works, real quick:

main: deletes the database
main: creates an empty database with one record and closes it
thread_a opens the database and sleeps for five seconds
thread_b opens the database and gets blocked waiting for a lock
thread_a wakes up from the five second sleep
thread_a writes a bunch of records to the database
thread_a dumps the database to foo02.db.dump01
thread_a gives up the lock and exists
thread_b wakes up with the lock
thread_b dumps the database to foo02.db.dump02 (you many of the
records written by thread_a are not found in this dump)
thread_b writes it's own bunch of records
thread_b dumps the database to foo02.db.dump03 (this dump has 
all of the thread_b records and is missing many of the thread_b records)
thread_b exits
main: opens the database and dumps it to foo02.db.dump04 (which ends up 
having the same records as foo02.db.dump03 showing that the corruption 
is in the database file on this disk)

These "threads" are just implemented as their own processes in UNIX.