One of our production database (Relstorage with MySQL backend) grow up to 450GB. What was significant - same database packed to data.fs (internal serialized Zope format) had around 350M :).

So, one afternoon I run OPTIMIZE TABLE on biggest table to perform vacuum cleaning.

Operation took around 8 hours (really) and after it, two Zope instances (from 4) were disconnected from database. Zope restart and..

MySQL throws in logs:

Mar 16 01:21:06 plone3 mysqld: 110316  1:21:06 [ERROR] /usr/sbin/mysqld: Table 
  './dbname/object_ref' is marked as crashed and last (automatic?) repair failed
Mar 16 01:21:06 plone3 mysqld: 110316  1:21:06 [ERROR] /usr/sbin/mysqld: Table 
  './dbname/object_refs_added' is marked as crashed and last (automatic?) repair failed

Hm, looks awful and dangerous. How it affects site? Plone works in read mode - users are not able to create content types. Quick look on Google doesn't help. So, let's check what MySQL can offer in this situation. I run on both tables:

REPAIR TABLE object_ref;
REPAIR TABLE object_refs_added;

It took around 15 minutes (for both together) to finish operation.

Zope restart - site working flawlessly.