When the system tests start up a Kafka Connect standalone or distributed worker, the utility starts the process, and if the worker does not start up within 30 seconds the utility considers it a failure and stops everything.
This is often sufficient when running the system tests against the source code, as the CLASSPATH for Connect includes only the Kafka Connect runtime JARs (in addition to all of the connector dirs). However, when running the system tests against the packaged form of Kafka, the CLASSPATH for Connect includes all of the Apache Kafka JARs (in addition to all of the connector dirs). This increases the total number of JARs that have to be scanned by almost 75% and increases the time required to scan all of the JARs nearly doubles from ~14sec to ~26sec. (Some of the additional JARs are likely larger and take longer to scan than those JARs in Connect or the connectors.)
As a result, the 30 second timeout is often not quite sufficient for the Connect system test utility and should be increased to 60 seconds. This shouldn't noticeably increase the time of most system tests, since 30 seconds was nearly sufficient anyway; it will increase the duration of the tests where does fail to start, but that ideally won't happen much.