Victor
Since I still am convinced that my 1015 has 256k of RAM in it I will try to explain exactly the problem I have.
1 Starting with a 1020 (with 256k RAM) - and according to the FlexRAM application note - its default bank mapping (with default fuses) is as follows:

which means 64k ITC, 64k DTC and 128 OCR.
So I start my boot loader with its stack pointer towards the top of OCR (in bank 7).
The boot loader sees that the application that I want to run is between 64k and 96k in size and allocates 96k to ITC, 160k to DTC and 0 to OCR and reconfigures the memory for the application - which now looks like this:

This allows optimal program and data access (zero wait state sand no caching involved or needed) but there is another important point of the strategy. The last FlexRAM bank 7 is always used for by the main stack pointer, meaning that its data is retained and the application can post messages to be boot loader by writing these just above the stack pointer at the top of bank 7.
If I use the same strategy on my 1015 it works exactly as when I use it on my 1020. The boot loader can locate its stack at the top of OCR and the application can communicate with the boot loader.
2. Now I us my 1010, which really only has 128k RAM and the strategy is slightly different:

It has 4 banks (instead of 8) and the default layout (default fuse setting) is for 64k OCR, 32k DTC and 32k ITC.
However the last RAM bank is not in OCR but in ITC.
This means that the 1010 boot loader uses the ITC for its stack. Assuming it needs to start an application between 64k and 96k in size it reconfigures the flex RAM as follows:

The application code runs again in ITC and has 32k for its data (and stack) so that optimal speed is achieved.
Bank 3 always contains the stack and so the application and boot loader can communicate with each other.
Now the problem.
I can configure my 1015 to do exactly the same as the 1010 does - to use only 128k (4 flex RAM banks) and the boot loader and applications do run.
But they can no longer communicate with each other due to the fact that the 4 banks are also not ordered in their default use in the same way as the 1010. The boot loader and application have their stack pointers (and communication mailbox which is persistent across resets) in different flex RAM banks.
Since I can physically run programs that use more RAM that the device theoretically has and also it does start with 128k OCR (which shouldn't be possible according to its data sheet) it behaves exactly as a 1020's flex RAM does.
3. For completeness I can show the way that my 1050, 1060 and 1064 boards work (with 16 flex RAM banks), whereby their application/boot loader communication operates correctly since all matches with the data sheets (and flexRAM application note)

Assuming the application needs 352k ITC for its code the boot loader allocates this to ITC and 160k to DTC, whereby the main stacks (and communication mail box) are always in bank 15.

For conclusion, I expect that the 1015 has the same FlexRAM default as the 1010 (application note):

but practically it has the same as 1020 (application note):

which is only possile when it has 8 banks and thus 256k.
All of my tests show this to be the case! Since its default bank layout is also not as documented it is furthermore not possible to use just 4 banks and 128k with the general boot loader concept.
Regards
Mark