Just to follow up on this because I think it's interesting =)

One possible utopian future is thus:

o Modify Koji to create profile-gathering versions of some targeted applications like Firefox; this would be a separate repository with packages of the same name
o Have a yum plugin that allows you to replace your installed packages with the profiler-compiled packages
o When you enable this plugin, Hotspot also goes into "dump profile information" mode
o A new Fedora server where people can post the data from their profiled packages and Hotspot logs
o Teach Koji to pull the data from this server and compile the software using it

To do this it might help if we only had one code generation backend for both GCC and Hotspot that understood how to process the profiling data - a long term vision on that would be to move to GCC-LLVM and Hotspot's Shark work.

This infrastructure would also help us make the decision which Java software makes sense to AOT compile versus not, rather than creating -precompiled versions of all packages.