Introduction
The optimization is about the animation node Blendlist_Base, which means all blend list nodes(eg. Blend Pose By Bool, Blend Pose By Enum and Blend Pose By Integer) can benefit from this optimization.

Performance Waste
The key point of this optimization is, in most of the times, only one branch would be evaluated for a blendlist.

But in FAnimNode_BlendListBase::Evaluate_AnyThread, a FAnimationRuntime::BlendPosesTogether function would be called even only one branch is evaluated.
1 | for (int32 i = 0; i < PosesToEvaluate.Num(); ++i) |
As a result, BlendPose<ETransformBlendMode::Overwrite> would be called, and this leads to a per bone blend:

How To Fix
It is fairly easy to fix this issue: We simply Evaluate the branch to the Output context when only one branch is valid.
