+1 to all the stuff Plec has said, of course.
I will already know, when I'm finished with a mix, what its loudness potential is, and how good it is likely to end up sounding. If I'm not happy with the mix I'm sending off, I will not be happy with the master. Nothing in the past has overcome that rule.
I always mix with loudness enhancement on the stereo bus, because it makes the most sense to me. Most rock/metal mixes are going to end up smashed, so I want to be hearing it in as close to a final state as I can while I am building the thing. This way I also know what steady-state signals are likely to cause clipping, and as a result I can put measures in to prevent this. Gauging by my own rough mix/masters I can tell where the ME is likely to have problems. What they're ultimately getting is a version of the product you have with much fewer parameters to control. If you don't think the mix, at its heart, can hold up to the process of maximizing, there likely isn't going to be much that the ME can do apart from crafty gain staging into certain devices, and some minimal compression and EQ. Certainly much less control and ability to reduce these problems than you have in the mix.
Ultimately whenever I'm not totally happy with a mix, it's more likely to end up sounding screwed after mastering. It usually implies some misbalance in the mix, some overlapping frequencies, chaotic arrangements, poor tracking etc. All these things can eat up your headroom, and waste energy where it's not needed. I've made enough mistakes, and had enough successes to start getting a greater feel for where this stuff might happen. Even so, some elements end up outside of your control despite your understanding of what they are. Knowing the limitations really starts to factor in. The process filters down from the writing stage, all the way through tracking, mixing and finally mastering. If the songs are arranged in a good way, there will be less instrumental clashing and chaos to deal with. If the tracking is good, you won't have to over-process and run the risk of weird frequency build-ups. If the mixing is good, then the mastering engineer has his work cut out for him.
I'm still making these mistakes, and learning as I go. The main revelation in recent times has not been to recognize the mistakes, but rather to recognize when it is that I can't do anything about them. The understanding that a finished song is one collective effort. Everything has to go right, from first base, all the way through to the final master copy in order to facilitate a comfortable process throughout.