Not yet in production but yes it works well :)
In our case, tables with cdc enabled have a low write limit so memtables are rarely flushed ... To deal with this, we trigger an explicit nodetool flush every X minutes, commitLogs are transferred well into cdc_raw and are consumed by the connector.
You can also deal with the following commitLogs parameters to, low setting could speed up the flush and transfer of commitLog into cdc_raw:
commitlog_total_space_in_mb, default 8G
commitlog_segment_size_in_mb, default 32mb