Thanks, Leon and David.
We (I'm working with another instructor) went with per submission so that the work would be distributed evenly amongst those in a group who submitted a paper to review (so a student who did not submit a paper would not be reviewing any papers). The groups (each group represents a tutorial of about 16 or 17 students; the course is set up in forced separate group mode) are separate to make it easier for the teaching assistants to assign participation marks for the reviews.
It's not a problem for a reviewer to get more than two reviews back on an assignment, but it would be a problem for a reviewer to only have one review to do. (Each reviewer submitted only one assignment.)
It sounds like everything is working as it should, then. By the way, this is our first term using Workshop - it's a great feature.