why would "shrink space casdcade" on a table cause hours of TM enqueue locking on that table?

 Database Enterprise Edition 10.2.0.3 on x86_64 RHEL4.

------------------------------------------------

I thought locking during an online shrink was supposed to be limited to row level locks (that is except for a table lock at the last moment when the high water mark is moved).  But that's not what we're seeing.  We're seeing a table get locked up with TM enqueues for hours, which is resulting in ridiculous wait times for any sql sessions that attempt to update the table.  We sampled a 3 hour period while the shrink was running, and results show the shrink session was responsible for over 2 hours of wait out of the 3 hour sample