We should recommend not to run more than 10 samples together because that will likely introduce artifacts. See the following quotes from the vignette describing the use of ExomeDepth.
ExomeDepth uses read depth data to call CNVs from exome sequencing experiments. A key idea is that the test exome should be compared to a matched aggregate reference set. This aggregate reference set should combine exomes from the same batch and it should also be optimized for each exome. It will certainly differ from one exome to the next.
Importantly, ExomeDepth assumes that the CNV of interest is absent from the aggregate reference set. Hence related individuals should be excluded from the aggregate reference. It also means that ExomeDepth can miss common CNVs, if the call is also present in the aggregate reference. ExomeDepth is really suited to detect rare CNV calls (typically for rare Mendelian disorder analysis).
The ideas used in this package are of course not specific to exome sequencing and could be applied to other targeted sequencing datasets, as long as they contain a sufficiently large number of exons to estimate the parameters (at least 20 genes, say, but probably more would be useful). Also note that PCR based enrichment studies are often not well suited for this type of read depth analysis. The reason is beacuse the number of cycles is often set to a high number in order to equalize the representation of each amplicon, which can discard the CNV information.