My 2p: My impression is that the typical peaks from FAIRE are too weak to be picked up by macs, so I agree macs is not ideal in this case. I got good results with fseq. "Good results" meaning the number of peaks was more or less as expected from the literature (tens or 100s of thousands, corresponding to ~1% of the genome) and from visual inspection the peaks looked ok. The method fseq implements seems reasonable and not overly complicated. Also, fseq runs ok without much tweaking of the parameters and without a huge amount of resources. In contrast, I couldn't get zinba to complete my analyses due to memory or time constraints or other problems with the underlying algorithm, but that could be me doing the wrong things.
I know, a lot of hand waving here. In general, I find peak calling to be very unstable. Depending on the peak caller you use or the exact parameters you get widely different results.