The common configuration property, query.fetch.size, uses a default value of 0 (zero). This default implies that the connector should use the driver's default value, which for Oracle is a relatively poor default of 10.
This property is currently only used for Oracle's LogMiner implementation when we fetch the changes from the V$LOGMNR_CONTENTS view. Based on the email thread https://groups.google.com/g/debezium/c/LaVR28b2QrQ, the user identified that the performance bottleneck they had was related to this configuration property and adjusting this value to something sensibly increased throughput.
I suggest we align this with the snapshot.fetch.size default value, which is currently 2000.
Additionally, we may want to review the documentation and determine whether a call-out to this configuration is warranted to help navigate users to setting this more often for performance if they're setting higher batch sizes than the defaults.