As part of our internal continuos performance testing, we have identified that acquisitions run around 25 to 30% slower compared to the legacy Autogetters implementation, that has implication on $lookup performance when executing locally on the classic engine. 

BM_acquire_ShardRole_stats/numSecondaryColls:0/iterations:100000       6457 ns         6455 ns       100000
BM_acquire_AutoGet/numSecondaryColls:0/iterations:100000               4847 ns         4847 ns       100000

As part of an extensive testing via micro-benchmark, we have not identified a single culprit, but several small changes can have an overall boost such as

• using intrusive_ptr instead of shared_ptr where applicable 
• move instead of copies where applicable
• avoid system calls where applicable