What is the best practice for mixing (or not mixing) different I/O load types in an aggregate? Should I keep similar loads together, or "mix and match" the sets.

For instance, I am looking at placing a variety of different data sets on four aggregates. These data sets include sequential backups-to-disk, nfs and cifs file shares, and "QA-dev" luns. Should I keep all the backups on one aggregate, and all the file shares on another? What do you generally do? Are there situations that will adversely affect performance?

TIA, Fred