The risk for illicit cooperation between bidders in procurement auctions may be particularly strong when many similar or identical objectives are to be allocated. This is so since the multitude of winning possibilities provides potent economic incentives and since the repetitive nature of these auctions provides opportunities for punishment of deviators. Moreover, this environment may also mean that bidders´ costs are non-linear in the number of contracts awarded. This paper presents the outcome of a wind tunnel type of experiment where three different mechanisms are compared. A standard one shot sealed bid procurement auction for two identical goods provides the benchmark. The same mechanism is then used to allocate goods with scale economies and finally a combinatorial auction is applied on the situation with non-linear costs. All three mechanisms are first run without, and then with the possibility for subjects to communicate prior to bidding over a chat line. There are two human and one computer bidder in each period. It is demonstrated that the combinatorial mechanism is able to improve efficiency. Subjects are also less inclined or able to cooperate under combinatorial auction than under the standard bidding format. The paper therefore provides indications of that the possibility to submit combination bids may throw some sand into collusive schemes.