The incremental analysis feature will be working fine in the partially distributed setup, but it wasn't thoroughly tested in the external hadoop cluster, hence it was marked as 'experimenal'. Anyhow if you find any bugs on these you can report it in jira.
To answer your question, you need to enable the incremental processing for your stream first and then you need to add the incremental annotation.The following are the detailed steps for this.
1) You need add property 'streams.definitions.defn1.enableIncrementalIndex=true' in the streams.properties as explained here file and create a toolbox which consists only the stream definition artefact as explained here.
2) Install the toolbox - This will register the stream definition you mentioned in the toolbox with incremental analysis. On this point on wards the incoming data will be incrementally processed.
3) Now indicate the @Incremental annotation in the query. The first iteration will consider the whole available data as you have enabled the incremental analysis in the middle of the processing, but from next iteration onwards it'll only consider the new bunch of data.