The problem with sky is that you're going to want multiple layers in your scene (assuming there are things that the player can walk behind) and for the 3D player to be able to change layers relative to the 2D layers.

View entities are the way to go for this type of system as they can freely move above and below panel layers. The usual 3D level is run invisibly to provide collision detection. (The view entity positions are updated from the invisible level.) [I take it that this is basically the case for games like Final Fantasy 7 and 9.]

The advantage of using view entities over having the 3D level change layers is that each character can change layers independently (say, if you have one character in front of a wall and another behind it). The 3D level would treat all characters as though they were on the same layer and therefore as if they were at the same depth in your 2.5D environment.

If I understand, the difficulty that you are facing is in getting the collisions from the 3D level (the placement of walls, for example) to match the visual of the wall in the 2D image. This should be somewhat easier using the above approach. If you're making all of your backgrounds in a 3D application, this isn't too bad, as you have the 3-space coordinates of all of your scene elements. And if you're drawing all of your backgrounds, you're going to have to painstakingly match everything by hand anyway.

In the style of game that I am going for, it is important to relay depth information to the view entities (so that one can run to/away from the camera in addition to up/down/side), but there are times when a 2D collision-detection system for view entities would be helpful as well.

Here is where Conitec would have to help us out. Implementing this would likely involve having a 2D image mark the places where the view entity should not be passable. The image would scale as the entity moves to/away from the camera. There would be a similar scheme for panels, and the two would interact to create collisions. If you drew all of your background images, then, you would simply draw lines over them to mark impassible regions for the character. (And Gamestudio would gain a valuable tool for making more conventional 2D games in the process.)