In your situation, suppose you think the criterion of 1bp reciprocal overlap is too loose, you can use -f=0.5 (in bedtools intersect) to keep each other have 50% overlap. However, as you know, CNV usually would be long range genomic variation, I think you can apply two stage screening: 1) use 1bp reciprocal overlap to find all the overlap event (CNV hotspot) and then 2) use strict criterion, maybe -f=0.5 or -f=0.8 to identify most similar or homogenous CNV regions. On the other side, the genomic location of the CNV would be very important, if they occurred in enhance region, then I think even 1bp reciprocal overlap would bring similar biological significance. if they occurred in intergenic region, then I think even -f=50% reciprocal overlap would be better to define same CNV event. In conclusion, you should dig the data deeply and show the relationship among the event comprehensively and to find the most biological relevance to the CNVs.

In case of CNV detection not every tool allows you to put in similar parameters for a calling the regions, however if you can put the parameters close enough like say the number of reads to be selected for a call for a tumor and a normal and the number of windows for binning or segmentation, then one can easily try to compare the CNV calls first based on the -f parameter of 1.-0(100%) which will most likely be less owing to its statistical model , but then then relax is to .8 (80%) or .5(50%) with bed tools. Finally when you have the similar regions you can in fact try to plot the correlation of the segment ratio score between the overlapping region to find the similarity for the regions calls.