I would recommend MACS2 with defaults (appropriate to your genome) as there is a linear normalisation of reads between the two samples, effectively bringing down the more numerous sample to the lower sample. Another important consideration is the genome you are dealing with and the number of reads that MACS will finally use (non-redundant reads). For a human/mouse transcription factor the general advice is 20 million, but sometimes you can get away with less depending on the binding profile.