Anyway, the problem seems to come from the fact that when I put the processor in sleep, the FTDI is not powered at all. Something very weird happens in that time periods inside the FTDI. When I plug the USB cable and the FTDI gets power, the consumption is much lower. That's why I believe that the FTDI is the source of the 'problem'.
Moreover, I have noticed a behaviour like that with a flash memory. The current consumption when the device was powered-down, caused by current flowing through the pins of the uC, was higher than the case where the device was in power. It may sound weird, but I have seen that in practise.
And have in mind that indeed I have removed the FTDI and the consumption falls to the levels that I mentioned. I also have another design with the same uC that can achieve that levels of power consumption (~200uA).
I completely agree with what you mention lower in your post. That 200uA is totaly higher than the number stated in the datasheet of the processor.
Perhaps my idea of putting some buffering or switches between the two ICs is not very conventional, but I am trying to think of a way to reduce the consumption as much as possible.