Never being one to ignore a juicy challenge, I submit the following examples of good design:
(1) The aforementioned PLATO chemistry course. In ugly old MPAS (the microcomputer derivative of Tutor). Designed about 15 years ago. Why do I say it's good?
(a) the presentation/practice sequences show a clear and tight correspondence to the knowledge structures--not just a bunch of isolated sub-tasks derived from a behavioral task analysis and sequenced with a learning hierarchy. All the more amazing, given when it was designed.
(b) meaningful interactions which are (mostly) appropriate for the type of learning intended. That includes a number of truly elegant and simple simulations, with great context-sensitive wrong-answer feedback. All implemented by a bunch of gifted programmers who were too dumb to quit (remember, this is MPAS/Tutor we're talking about!).
(c) an engaging, light style throughout which conveys the intrigue and fun of science, and never talks down to the learner or patronizes him/her.
(d) carefully laid out screens with good visualization and text design, given the extreme limitations of the available technology.
(e) objectives, presentations, practices, and mastery tests which all align in content and type of performance required.
(2) An air traffic controller instructional simulation. Has these features:
(a) high cognitive fidelity to the task, and enough "physical" fidelity to the actual data displays and other cues so transfer is very likely.
(b) excellent integration of declarative knowledge teaching (through introductory tutorials wrapped around the simulation), using design practices Dave Merrill and Charlie Reigeluth would be proud of. (In my experience, many problem solving simulations simply ignore the acquisition of declarative knowledge.)
(c) very good, context-sensitive, cognitive coaching at the strategy level. Clearly based on a thorough cognitive task analysis of the entire performance environment and individual tasks.
(d) lousy mental modeling and no mental modeling practice except for the simulation. In my view, they blew it on this one. They probably got away with it in their evaluations because the test population was clearly experienced novice-level controllers. But still...
(e) all runs from a multimedia CD-ROM on a standard Windows machine. Done in a combination of C++ and Toolbook, if I remember correctly.
(3) (Claiming developer's privilege here) the projects we're working on right now (by definition, any Instructional Developer's really good designs are in his/her next project...). Lots of neat problem-solving instructional simulations, doing some things that haven't been tried on a "mass production" commercial basis before (if we can get it all to run...). I could tell you more, but then I'd have to kill you...