Some people have suggested that the almost complete overlap obtained with 2x250 V4 is a good strategy because it gives the best error correction. I disagree. In my opinion, for most applications the reduction in error rate is not dramatic and does not compensate for the improved phylogenetic resolution you get with longer sequences.
Using UPARSE or UNOISE gives very accurate OTUs regardless of the overlap length. As a proof of concept, you can generally get good OTUs if you throw away forward reads for V4 and use reverse (R2) reads only, despite the lower quality of the R2s. I therefore recommend sequencing the longest possible amplicon that will give a minimum overlap of around 20 to 30 nt.