Reason missing for existence of pooling/connection_cache setting

[tengcomplex]

The documentation just mentions the possibility to disable it, without deeper reasons.
Looking at source it's implemented for a long time.

Questions, are spotty networks the reason why this option exists?
Am I right to assume it's a little more stable running Pgpool without connection-cache for the price of a little less performance?

[tatsuo-ishii]

The reason for disabling connection cache for regression database is, to run PostgreSQL's regression tests. The test runs tests in parallel. If connection to regression database is cached, the test fails with:

ERROR:  database "regression" is being accessed by other users
DETAIL:  There is 1 prepared transaction using the database.

The reason for disabling template0 and template1 is closer to your guess:

Am I right to assume it's a little more stable running Pgpool without connection-cache for the price of a little less performance?

Applications/tools using those databases are often critical ones such as createdb and dropdb. For those tools stability is stability is the top priority IMO. Moreover connection pooling does not give a lot benefit since those tools do not repeat connecting/disconnecting to a database.