We have built a collaboration / business network tool for a specific niche.
So we would like to introduce a recommendation engine that will suggest which users to connect to.
All users are suppliers and consumers at the same time. Users that supply similar items are competitors and would have no interest in collaborating. Users that have demand for items another user supplies are good fit.
So the ideal scenario is not to do this based on similarity but on how well 2 users complement each other in terms of what they supply and what they have a demand for.
Is this a good use case for mahout or is the whole concept based on similarity?
While you could adapt a recommender approach to do something meaningful here, from your description, this does not sound like a recommender problem. You are trying to recommender users to users, but are not basing it on any user-user interaction. I also think there is little room for serendipity or near-matching here: if I need an item, I want people that can supply that item, not things kind of like it.
I wouldn't overthink it. Use the simple approach until you know it won't work for you. Just match users those supplying the most different things the user needs (or vice versa).