There is a known feature request to improve the feature related to this (e.g. do not allow users to complete the workshop unless they really provide and receive the expected number of assessments). In the current version, teacher have to check these things manually.

The current behaviour is that if the peer does not finish the allocated assessment (such as Toni Tre does not provide assessment of Toni Due's work in your screenshot), the allocation if ignored. So as you say, the final grade is calcaulated based on the 4 submitted assessments. Workshop indeed distinguishes between grade 0 (= very poor submission) versus null / empty grade (not assessed).

grade is enhanced by mistake, because if Toni Tre's reviews was 0, the final grade was 13.

If this were the real situation, I would probably go and check the peer assessment by Giuseppe. If his assessment is right (and all three others are wrong), then I would increase its weight. If Giuseppe's assessment is wrong and all three others are right, I would lower the weight to 0. Then, after recalculating grades, Toni Due the author as well as all the reviewers should receive more fair grade.