Mark,
While "sims" in your comments is correct that it does depend on the source material, and I've never met two professional mastering engineers with the same signal paths, I will give a shot at trying to answer your question based on my own mastering work.
Generally my chain ends up in blocks, in the following order, though any block may be in or out at any given time:
- Minimum Phase EQ
- Linear Phase dynamics Pre EQ
- Dynamics blocks
- Linear Phase EQ - post dynamics
- Ambience
- Stereo width
- Final eq
- Limiting
I also tend to use a good amount of mid-side processing (see my blog for a long primer on mid-side)
To answer your specific questions:
a) I'll use eq where ever I feel like it, and almost always on either side of my dynamics to pre and post shape the audio into the dynamics to get them to do what I desire.
b) I tend to use multi-band compression as little as possible. Generally the better the incoming mix, the less I use. Multiband can be a useful bandaid for poor mix balance, but it's not a substitute for mixing chops. My most common use of multiband is typically a single band to fix a specific problem. For instance, if there is a particular frequency band that needs a lot of eq to fix balance issues, sometimes a multiband before that eq will smooth the band enough to make the eq correction less apparent.
c) Ambience with a reverb or delay is a rare, but sometimes used item. More common is tailored dynamics to shape the ambience in the original recording.
d) When I find a tape plugin that I like, then maybe.
e) I don't use a level metering plugin. First, many of the don't correctly apply a reconstruction filter, and therefore useless for detecting intersample overs. Second, because of the phase shift near nyquist in some implementations of the various lossy codecs that people use almost exclusively to listen to music, the final peak levels of the wav/aiff will not correlate to the final peak levels of the mp3/aac. Almost inevitably you have to print the lossless file a little lower in level, convert it, and then look for intersample overs in the resulting file, and adjust the print level of the final product to give as much as possible without those intersample overs.