What this attempts to do is remove false-positive apparent coverage increases due to PCR duplicates. If you happen to have paired-end reads, then you might be better off using something like Picard's MarkDuplicates tool and then just filtering reads out. How important this setting is will end up depending on how high your average coverage is. Unless you did a lot of sequencing, it's likely that apparent duplicates are actual duplicates, so removing them with this option makes sense.

BTW, you can likely see how big of a problem this is by making a histogram of per-base coverage with/without these reads. From that you can likely surmise how big of an issue PCR duplicates are in your dataset.

On other thing to keep in mind is that while removing these will decrease false-positives, it may also decrease true-positives. So how expensive/time consuming/sensitive your downstream analyses are may dictate how you treat these reads.

I have been using MEDIPS for over a year and after extensive comparisons of the impact of filtering out duplicates (via Picard's MarkDuplicates) or not, we decided it was best to filter them out. So, if you don't want to go through the trouble of doing that in Picard, you can use the uniq=TRUE setting in MEDIPS.