Linux (the kernel) has been aware of this result for a long time. Here's a thread from 2001, where Linus says "I would, for example, suspect that a "correct" o

Linux (the kernel) has been aware of this result for a long time. Here's a thread from 2001, where Linus says "I would, for example, suspect that a "correct" optimization strategy for 99% of all real-world cases (not benchmarks) is: if it doesn't have floating point, optimize for the smallest size possible. Never do loop unrolling or anything fancy like that. " http://gcc.gnu.org/ml/gcc/2001-07/msg01543.html http://gcc.gnu.org/ml/gcc/2001-07/msg01543.html and a similar thread http://lkml.indiana.edu/hypermail/linux/kernel/0302.0/1068.html http://lkml.indiana.edu/hypermail/linux/kernel/0302.0/1068.h... And the 2.6.15 (2005) changelog which exposes a configuration option to compile the kernel optimized for size http://lkml.org/lkml/2005/12/18/139 http://lkml.org/lkml/2005/12/18/139