Optimizing compiled binaries

Webfaction seems to heavily encourage (or at least support) users compiling their own packages. I got curious about the compiler optimizations that are enabled by default. echo $CFLAGS returned nothing. Reading make output looks like the default optimization setting for compiling is -O2. There seems to be no architecture-related optimization going on.

Over the last couple of days I've been playing with a new server stack and I did a side-by-side comparison with different CFLAGS. Compiling PHP with -Os instead of -O2 netted a 60% reduction in the binary size, from 30MB down to about 9.5MB. That was the most drastic change, but Nginx and uWSGI also showed demonstrable improvement in binary size from using -0s. I also added -march=native to my CFLAGS for the full set of architecture optimizations available.

Since WF doesn't seem to set these as defaults, I'm curious if there's any reason I shouldn't do it on WF servers. I can't detect any performance or stability hit from using -Os instead of -O2. It would seem like in an environment where RAM is at a premium on the lowest tier accounts, using the best memory optimization by default would be the most beneficial. If I'm not mistaken, when a package isn't compatible with a certain compiler optimization it will force it to be disabled. However, since -Os is just -O2 with size-increasing optimizations disabled it should have absolutely zero impact on compatibility.

Is there any reason I shouldn't be setting my own CFLAGS on Webfaction? If not, might I suggest setting CFLAGS="-Os -march=native" as a global env variable? -march=native has the downside of making a binary that is incompatible with other architectures but I don't think many people are compiling binaries to use on other servers via their Webfaction accounts so I don't see any practical cons.

Thoughts?