That's pretty much what I thought too.

So the real question is whether the commons pooling does this. Does
anybody know the answer to that?

Using Oracle JDBC connection caching with Tomcat should not be
different than using it with a stand-alone Java/JDBC code.
In summary:

a)create an Oracle data source
  OracleDataSource ods = new OracleDataSource();

b) set the DataSource property ConnectionCachingEnabled to true
  ods.setConnectionCachingEnabled(True);

c) optionally set more properties (see the Oracle JDBC doc for more
details @
http://www.oracle.com/pls/db102/to_pdf?pathname=java.102%2Fb14355.pdf&remark=por
tal+%28Application+development%29
)
  ods.setConnectionCacheName("MyCache"); // optional
  ods.setConnectionCacheProperties(cp); // optional

d) getConnection() will service connection requests from the cache (you
may pre-populate the cache with minimal number of connections,
otherwise the first connection requests will trigger connection
creation).
 conn = ods.getConnection();

For Fast Connection Failover in RAC environments using 10g JDBC:
1) you need to configure ONS so that JDBC can subscribe to RAC events
and be notified upon node/instance/service failure or the addition of
new instance.
2) Then, enable FastConnectionFailover either programmatically
     ods.setFastConnectionFailoverEnabled(true);
   or using  system property:   java -D
oracle.jdbc.FastConnectionFailover=true

3) catch fatal SQLEXception (and retry if so). fata SQLException is a
list of SQL exceptions related to node/instance/service/network down
(user can programmatically add it's own exceptions)

See codes samples @
http://www.oracle.com/technology/sample_code/tech/java/sqlj_jdbc/files/oracle10g
/index.html
See also complete step by step instructions in chapter 7 of my book @
http://www.amazon.com/gp/product/1555583296/

Kuassi, http://db360.blogspot.com/