First, I organize ALL my classes-- university biology/zoology/ecology classes, mostly sophomores through seniors-- around small groups. Most of the assignments are group assignments, as are portions of the exams. Students typically work with the same group mates through out the semester. I've been doing this for ten years and have tweaked course structures until they work very well indeed.
There is some data available about self selected verses externally selected group membership-- I don't have citations in hand but I'll refer you to Michaelsen et al Team-Based Learning: A Transformative Use of Small Groups, which discusses some of that data if I recall correctly. The short version is that effective group dynamics take time to develop and often go through predictable developmental stages, but preexisting relationships among group members slows down and sometimes completely disrupts the evolution of effective group learning strategies. For example, one of the biggest problems students often have is effectively challenging one another-- established friendships tend to make it even more difficult to challenge each other in discussions, study sessions, etc. They also lay down preexisting channels of communication and social protocols that can exclude other group members. Perceived exclusion can relegate some students to social loafing from the first day of group establishment.
I usually choose groups during lab sections, since students need to work in groups during lab (all of my classes are lab classes). I've found that people who have preexisting friendships usually seek one another's company during the first class sessions, so if there are, say 25 students in a lab section, I'll form them into five groups of five members by asking them to count from one to five as we go around the room, and all the ones' form a group, all the twos, and so on. In some classes group sizes vary, but never by more than one up or one down. Since friends tend to sit together, the counting off effectively randomizes their distribution among groups