As for loudness war, we're making a stock music for use in someone else's projects, not a music for a radio station, so I think it's completely irrelevant. Nobody will buy AJ tracks just to listen to them in a boombox after all.
My opinion is that we should provide a versatile recording that someone will be able to edit and mix with other audio material, speach tracks in particular. So there shouldn't be too much dynamics to make levelling with speach easier, but the customer may compress it further if they need, but it also shouldn't be too flat because the customer may want more dynamics in some applications. The same applies with EQ or other effects - it's easier to add them in production, than to get rid of them when they've already been aplied.
My first step of mastering chain actually starts in arrangement - I do everything I can to make everything sound right when played from the sequencer. This way I can get the best possible quality already in the recording.
As for such advanced effects as stereo enhancement, or heavier use of basic tools like EQ or compression, I apply them only to selected tracks before mixing, if it's really needed. Usually I put a rhytm guitar very wide in stereo image, and this is done on the individual track or group before the mix.
I keep my mixing simple, but I spend a lot of time listening to the mix on different equipment to ensure that it's working not only on high quality monitoring system. It takes days to make any adjustments and mix again and listen to it again on every possible low end player I've got, while walking on the street or working the day job.
After the mix is approved, there isn't much to fix already. The small EQ tweaks and a delicate compression to ensure a desired RMS level is all I need to do. And again, the most time needed for mastering is the time spent on listening on different equipments and adjusting the sound accordingly.