A continuous ground plane is important under all signals, but is absolute under any high speed signal. If you were using the ground plane layer for traces that was off to the side and your processor/memory was way on the other side, I'm sure you can get away with it. Two layer boards run without a ground plane, but you can't go fast on them, generally speaking.
10 layers could help, but I also think it is important to analyze the stack up of the board. I think that is the real source of your problems. I forget what speed the i.MX21 tops out at (266MHz? 400MHz?), but I am sure it is fast enough (clock speed, slew rates) that using impedance control on 8 layers would work just as well. Also, trace length matching might be important to re-examine if you are not careful there. I don't know if you were concerned about any of this when you made these boards?
Again, it's hard to judge but I think there was a project that uses the i.MX21 on six layers. I forget the product but it is an open sourced box that sits next to your bed like an alarm clock and it has an LCD. Maybe someone else knows it and if gerber files are available you could see what they did.