From @arulajmani's comment in #121088:

I was thinking we could add metrics/logs on the response path whenever tracking pipelined locks for a ranged request causes us to go over budget. If imprecisely predicting the memory usage for ranged requests ever becomes a problem, we can always reach for the cluster settings that disable lock pipelining for DeleteRange/locking {,Reverse}Scan requests.

Jira issue: CRDB-37280