Probably wouldn't work in the general case, as each of the bytecode streams would generate memory accesses that are completely independent. Not to restate the

Probably wouldn't work in the general case, as each of the bytecode streams would generate memory accesses that are completely independent. Not to restate the obvious, but SIMD relies on locality of data access to keep the SIMD unit fed with enough data bandwidth.