STAR multi-mapping

[id0]

By default, STAR outFilterMultimapNmax parameter (read alignments will be output only if the read maps fewer than this value) is set to 10. Should that just be set to 1 if you are going to process the data with htseq-count, which discards multi-mapped reads? Is there a benefit to leaving it at 10?

[alexdobin]

By default, STAR outFilterMultimapNmax parameter (read alignments will be output only if the read maps fewer than this value) is set to 10. Should that just be set to 1 if you are going to process the data with htseq-count, which discards multi-mapped reads? Is there a benefit to leaving it at 10?

Even if you keep multimappers in the SAM output, HTseq will not count them - it recognizes the unique mappers by NH:i:1 attribute.
Whether multimappers are useful depends on your downstream analysis. I find it useful to generate wiggle tracks separately for unique and multi-mappers, sometimes you can miss expression if you only look at unique mappers. Also, other quantification software (e.g. Cufflinks) can "rescue" multi-mappers. Typically multimappers do not occupy a lot of space, and - if needed - can be easily filtered by NH flag or the MAPQ field.