Firstly, 20 million records is not a large database as far as MySql is concerned.
Secondly, inventing your own half baked method of replication should be avoided completely. MySql has support for high availability and scalability. See http://dev.mysql.com/doc/refman/5.1/en/ha-overview.html
As for Google. They use there own database engine (BigTable) for most of there database work not MySql. There are open source (similarly designed) databases such as cassandra (developed, used and made OSS by facebook) but they are a completely different database model to that of MySql and use a NoSQL approach.