My armchair researching dug up a lot of articles about this. I'm not an education scholar but here's a brief slice of some results from some research in the business education literature.
From Randall S. Hansen (2006): Benefits and Problems With Student Teams: Suggestions for Improving Team Projects, Journal of Education for Business, 82:1, 11-19
In their related work section, their works surveyed suggest that professor-selected groups appear to have a more positive experience.
However, Muller (1989) stated that student preferences are not
necessarily the most important criterion for successful group work,
whereas Koppenhaver and Shrader (2003) suggested that
instructor-assigned teams lead to more stability in membership, and
that stability enhances each team’s ability to perform effectively.
Contrary to earlier researchers, Hernandez (2002) stated that student
teams should be formed by the instructor, and that students are more
likely to have a positive learning experience when groups are selected
by the professor.
In an empirical study by Praveen Aggarwal and Connie L. O'Brien (2008): Social Loafing on Group Projects: Structural Antecedents and Effect on Student Satisfaction, Journal of Marketing Education. 30:255, they hypothesize that self-selection of groups might reduce social loafing, based on some related work below:
[...] This prediction has some precedence in the pedagogical
literature. Groups formed by the students instead of random assignment
by the instructor are assumed to be more cohe- sive, more productive,
and experience a lower incidence of social loafing (Strong & Anderson,
1990). Mahenthiran and Rouse (2000) found that paired groups of
friends had less incidence of social loafing than randomly assigned
However, after their empirical study of 420 students in marketing and marketing-related fields, they conclude that there's no effect of student self-selected teams on social loafing.
In this study we proposed four such factors: reducing the scope of the
project, reducing group size, allowing students to self-select group
members, and including multiple peer evaluations. Three of the four
factors were found to have an impact on social loafing.
The only one that didn't have an impact was self-selection.
So one paper says that self-selection is worse, another says it generally doesn't matter. This might suggest that there are other factors that you may want to consider first with respect to making teams perform.
However, as mentioned, I'm not a business education researcher, and these projects might not generalize to other domains either. For example, the fact that these two papers don't cite a similar body of work might suggest that there's a lot more stuff out there. If someone out there is able to expand on this that would be excellent.