We can afford spending more compile-time in FTL, and we have more non-local optimizations likely to benefit from lots of inlining.