Mentoring students does take work, some do try and game the system. That happens no matter if GSoC or not. With GSoC, you basically have a form of financial support for hiring students. The GSoC paperwork is also rather minimal.
If you can, going through local universities and colleges is also a great way to have students work on community-driven Free Software projects. On a project, we had 4 engineering students do the equivalent of an internship on our community mesh network project. They did really great work and it's nice to have ties with university communities. We also had profs in social science classes inviting us to give talks, or had networking workshops during the weekends.
IIRC, the student gets half the payment on completion of the first half of the project, and half after the second, both on the condition of positive reviews from the org. So if a student just blows it off, they don't get their payment.
I think a better system would be some kind of endorsement program, where Google donates money to projects and in return the projects agree to mentor one or more student contributors. At the end of the summer if everything works out, Google can endorse the student for the work they've done. The project gets something done, the extra cost of bureaucracy is covered by Google, and the student gets real world experience verified and endorsed by Google.
You seem to draw a conclusion solely based upon GSOC not being a good fit for the OpenBSD devs.
The first thing that comes to mind is pay a stipend to the project and the student. So if the student gets $6K and 99.99% of them skip town at the end of summer no further contact, the project gets nothing but unmaintained code and wasted effort. However... what if the project ALSO got $6K of cold hard cash? Or credit for google services, adwords, donated hardware, who knows...