Build times when using the quick flavour:
stage1 opt | time (wall) | time (user)
-O1 | 13m | 53m -O2 | 13m | 51m
So even when we compile stage2 with -O0 (quick)
using -O2 on stage1 is already faster.
The difference is even bigger when freezing
stage1 and doing multiple builds or compiling
stage2 with optimizations.