Going back to the deficit, there is no science here. To have a positive contribution one should do a few more reviews than one receives from own submissions. To be neutral, one should do at least the same number. This would be equivalent to the above privileged situation of simply posting to arXiv and don't do any review at all.
If you work with coauthors, and everyone takes the attitude of "one review for one submission," then there is a net gain in the system.
Also, sometimes I get asked to review I paper I'm interesting in reading deeply anyway, and these reviews are the most pleasant.