I would suggest using mitochondrial cytochrom oxydase I (Cox I or CO I),
although it does not fit ALL your criteria. But:
1. Its Folmer fragment is read for enormous amount of invertebrate taxa,
so you may design model datasets ranging from inter-type to populations
2. Base composition there is peculiarly biased, which underlines the
necessity of very carefull choice of substitution model:
the gene resides at h-strand, but 1st and 2nd positions are G-enriched due
to AA composition, thus 3d pos appears "super-biased" to compensate that.
3. Effects of codon usage there are really dramatic
4. Structure & function are known, and its fun to see AA substitution
frequencies along the sequence. Interestingly, they differ in distant
groups of organisms, so there is very illustrative to trace the
differences and map them to the structure which is relatively simple
(transmembrane alpha helixes/loops).
Acc. to my experience (only 1 year long ;), students freshly out of
invertenrate zoology enjoy playing with it during practicals.