I add 1 frame of delay into the input system of a game every time someone writes "2.5D".
Isometric beats 3/4 view for communicating 3D. I've shown isometric screenshots and had people say "ah! 3D!". Issue is, isometric view can fail if you have geometry that the player can walk behind - the top of the geometry might blend with the floor level the player is at.
3/4 view has this issue too, but gains the benefit of having more artistic freedom to better communicate what's going on (it is already an illogical view perspective, so can get away with trading logic for readability).
The big problem people hit when trying to communicate height in 3/4 view is the Y axis on the screen. Walking up the screen does nothing at all to communicate height change. The SNES Final Fantasy games do a good job with cliffs and mountains; these games make the player walk left/right on the cliff, with thin paths, but tall mountain walls. In Final Fantasy V and VI you had to enter caves if you wanted to go up to the next level of the mountain (so the player would almost never walk "upward" on the screen Y axis). Add in a pretty parallax picture and then you've got easy mountain/cliffs. The Celes suicide scene in FF6 is side-on.
Actually now that I think about it, even the buildings in FF6's Zozo force you to only really move side to side (or back inside the building), so even these go "side-on" for conveying height. Same deal with the mage tower. So if you want to show someone jumping and changing height, don't do it up/down on the Y axis - always do it sideways (diving into water? Make them dive to the left/right). Phantom train was side-on also, with a horizon. I bet the train would look rather flat and not tall at all if someone stuck it in a normal map.
Sword of Mana on the GBA comes to mind as one of the worst games for conveying height changes. Jumping onto ledges felt like a guessing game (is this really a ledge or is it land I can't reach?). Spent so much time jumping at walls just in case a ledge was reachable.
Graphically, point of references are good for height. Put a tall tree at the bottom of a cliff to show that the wall is definitely "going up".
I would guess most people would just stack stuff on the Z layer (into the screen). Have a parallax map that shows a tiny town in the distance from top-down perspective (so no horizon visible) or dim the lower layers if you change height - I consider these lazy and not very effective. Have a horizon, have it side-on.
I really like the idea of zooming the camera through clouds. Maybe that could work for trees in a forest somehow. That could work for diving into water too - zoom the camera into the water and fade to black.