See QA-10

4) It looks like we only check that the beginning of a record is in memory, not the end. So if we fall back to using process info to see if a record is in memory we might get a lot of false positives about records being in memory when only the beginning of the record is - particularly if we are reading back to back medium to large size docs and a doc seeds the beginning of the following doc into memory.

6) It looks like when we are calling touch() we currently just touch the beginning of a record not the whole thing.

• delete {} - in this case the new code was better on average. With the new code time to acquire a lock was still a lot slower running against deletes than running against nothing - potentially running into the issue described in 4) above