hm. so, definitely for cold compiles. less clear what the speedup would be for incremental compiles