Use case: Your language allows users to define new packed struct types at runtime. Your language toolchain targets wasm/gc. You use an (array i8) to represent the backing store for those packed structs.

Problem: For multi-byte loads you have to emit multiple array.get_s or array.get_u calls and then combine the bytes appropriately. This is inefficient, which is enough of a motivation to add multibyte accessors. There should be something to load and store u16 and i16 from arbitrary offsets in an (array i8), as well as i32, i64, f32, f64, and i128.

However I just realized another motivation for multibyte accessors: byte-by-byte access is potentially incorrect in the presence of threads and mutation. Unlike naturally-aligned access to memory in the MVP, access to (array i8) contents with MVP GC ops will tear. Not sure what to do about that: whether to specify that naturally aligned multibyte access does not tear (perhaps with an exception for i128), or whether to ensure atomic access only via specific atomic operations. In any case there is some design work to do here.