The use cases for query-only database replay are:
To warm up the database buffer cache
In some cases, a workload is captured when the database buffer cache is warm (data blocks are already in the buffer cache). However, when you replay that workload on the test system, the buffer cache will not be warm, and the data blocks will need to be loaded from disk initially. This may make the replay duration longer than the capture duration, and increase the database time.
To avoid having to warm up the buffer cache, you can perform a query-only replay and then perform the read/write replay without restarting the database and without flushing the buffer cache. Note that you do not have to restart the database after a query-only replay because a query-only replay is read-only.
To find regressions
A query-only replay is a good and easy way to find regressions from the read-only part of the workload with concurrency. The read-only part includes SELECT (not SELECT...FOR UPDATE) statements, PL/SQL without DMLs and DDLs, LOB reads, and so on. It is typically the main part of the workload capture.