You've gotten suggestions that can surely help, such as locking the lock table. That type of solution can have significant problems when some process dies unexpectedly and the lock isn't removed. Perhaps you'll have a cleanup script that removes dead locks or some such.

How about considering implementing something like common version control systems use, Optimistic_concurrency_control (or "optimistic locking" which Wikipedia tells me is a misnomer). Time stamp or version updates to data; when updates conflict, ask the user creating the conflict to fix it. The idea being that conflicts are uncommon.

Only you know your application, so I may be talking out of my hat. However, you may find you're going to extreme ends to account for a circumstance that rarely occurs.