Could you please elaborate why it is a better choice? I don't see how django-beanstalkd could take much less memory than celeryd. If it's about the broker (beanstalk), you could have implemented a beanstalk backend for celery as well, then you would have had a tested implementation of the worker part.

Btw, If you would be interested in writing a beanstalkd backend for celery I would be happy to help you out!