> 1) I see some papers reporting the angle at which axial images are acquired. Why is this angle non-zero?
> 2) Why is the plane (orientation) in which images are acquired such an important parameter of an MRI scan

Perhaps you are referring to the flip angle? The angle of the volume orientation is another angle, but I've seldomly seen it reported as a number (usually just a description, like slice were approximately aligned to this-or-that brain structure). The flip angle refers to the spin physics and plays a major role in the contrast of the image. This one must be non-zero for an RF-excitation to have any effect.
If you do mean orientation angle, then it doesn't seem overly important indeed. Unless you have non-isotropic resolution (different in-plne resolution from slice thickness), in which case it may be useful to know in which direction data are how accurate. Also, when judging artifacts (from scanner, or from interpolation), the orientation is relevant of course.

> 3) What is the safest thing to assume when papers report coordinates but do not specify A. whether they are in voxels or in mm B. whether they represent the coordinates of the centre-of-gravity (i.e. the stats-weighted mean) or those of the maximum (peak) activation?

Coordinates are virtually always in MNI space if these are normalised data (sometimes Talairach, which is comparable but not identical). This corresponds with voxels or mm if you have a 1mm resolution normalised template, but better just see it as dimensionless numbers in a standard coordinate space. Whether it concerns peaks or centre-of-mass should hopefully follow from the paper; if nothing is said, most likely it will be a peak coordinate (the largest peak in the connected blob, if there is only one reported).

> 4) Why is it that some papers quantify BOLD signal strength (activation) as a t-stat whereas others report a z-stat?

It is always a measure of significance, that is the ratio of effect size, i.e. fMRI response amplitude, relative to the standard error of the same. The t-value really is like a z-value, except that it accounts for the fact that the effect size and standard error were determined from the same (small) sample, in which case the shape of the distribution changes slightly. You can convert t-values to p-values, and those to z-values if you like. Conceptually, t and z is quite comparable, and for large numbers of degrees of freedom, even the numbers are alike. I don't have any idea why some people feel a need to convert to z if the original result was t.

And then your most interesting question, if you ask me: ;)
> 5) Why is activation strength quantified in terms of significance (it seems most heatmaps are based on p- or t-values) and not on some measure of effect size, which seems to me would better justify the term "strength of activation"?

I agree! Both are different ways of stating something about "strength". The betas themselves, preferably normalised to the baseline level and expressed as a percentage signal change, quantify the magnitude of the effect in the brain. It does depend on the imaging sequence, so it still isn't intrinsically "meaningful" as a biological measure, but it allows you to compare more easily across studies with similar paradigms. The t/p/F-value says someting about the significance of the effect, so how confident you can be the effect is real (or much better: how unlikely the observation is if the effect were zero!). Significance depends on sample size, effect size doesn't. Sometimes one is more interesting (when trying to reject some theory as not fitting with the data, for instance) and sometimes the other is more interesting (when trying to get some idea how physiologically/clinically relevant an effect seems to be, for instance). Ideally, report on both; effect sizes are reported far too little, if you ask me.

Dave