I personally do not think LS on physical clusters should be maintained any longer. It is enormously complex solution which brings way too many limitations to be viable these days.
Just a couple if examples:
1. In Unicast LS mode, with two members, pivot takes care of 30% and forwards other 70% to another member. In terms of bandwidth it means you lose up to 40% of wire speed against HA, just because of sending the same packet twice.
2. With Multicas mode, Decision Function and Flask&Ack sync mean more latency and potential degradation of PPS against a regular HA, especially if you add more cluster members. Not even talking about acceleration drawbacks.
LS only made sense with slow FW inspection and low speed interfaces. Neither is the case today.