The CD audio works fine for me, and I use the same method of mounting the images as you do. I believe they are recorded as long audio tracks with multiple conversations, and the game then seeks to a particular point depending on what's being said. So by the sounds of it your cue sheet is being created with a 3 second offset for some reason.
I have a different problem though which you might be able to check for me. I mount the images the same way you do, using the multiple image mount command and using ctrl+F4 to switch between them. However, if the wrong CD is inserted (just mount CD 2 and talk to someone in the first town) a prompt is shown to switch CDs, but it just gets stuck there for a while then quits out. I'm not sure if this is because the internal image reader doesn't send a prompt to the game telling it a new CD has been inserted, or if it's something else to do with the game itself.
I'll have a look at my .cue files when I get home from work and post the proper offsets (if they're there!) to see if that helps you.