It was my understanding that apps without special x86-only inlined assembly would be straightforward to port. This could hint that LR, as slow as it is, actually has special vectorized or otherwise performance-boosted sections of code.