Unstable benchmarks should be removed from the system:
1) Avoid building/running them in the first place (use the "make check trusted" mechanism under tests/benchmarks for this)
2) Delete them from the database
Note that this task represents a permanent activity: The set of benchmarks changes over time, and new, unstable ones may pop up. Moreover, a particular benchmark could also change its own stability characteristics over time (in theory at least).