You need to be aware of performance tail. If you run 8 jobs each with 1 parallel stream and 6 complete quickly, you will have a tail of 2 long running single stream jobs.
It will likely be better to run each single job with 8 parallel streams to ensure parallelism all the way through.
We also mix and match based on expected runtime with shorter jobs getting single streams and longer jobs getting parallel streams.
Of course, you have to ensure parallel streams are well tuned. Is 2 really faster than 1?
Generally speaking multi-stream backups are best for backing up large objects since multi-stream backups will split the object across all of the streams and backup/restore them in parallel.
At a very high level (depends on a bunch of other factors) assuming an MPP solution if you had 8x tables all the same size and shape and backing them up with 1x 8 stream backup would be faster than 8x 1 stream jobs due to some internode processing (refering to Enhanced Local AMP backups).
nb. unlike backup if you need run concurrent restore jobs you will need to have seperate restore accouts for each
Again in real life where there are other technical and non-technical obsticals it may make sense to cut up some jobs and run concurrent backup jobs.