I am working on updating the Runtime Mesh Component plugin for UE5. This plugin works with mesh data buffers, and many of its blueprintable functions use TArray<FVector> and TArray<FVector2D>. These buffers are internally 32-bit floats, which matched the internal representation of the FVector types pre-LWC.
With LWC, there is a significant overhead introduced in UFUNCTIONs that need to use TArray<FVector3f> internally, because only (default double-precision) FVector is supported as a function argument and conversion is required. Maybe there should be Blueprint support for single-precision floating point types to improve performance and memory footprint?