I found some errors in your included instructions

FPCMake modification
--------------------
Copy the files you found in "FPCMake" directory of this package in the directory

%FreePascal%\source\utils\fpcm

In msys (or a dos prompt), go to %FreePascal%\source\utils\fpcm and run
"make". Go to %FreePascal%\bin\i386-win32 and copy here the new generated file
fpcmake.exe. This utility is useful when you try to build the rtl, because it
generates all makefiles starting from a smuch more simple makefile.fpc.
Running 'make' does not produce a single exe file. Instead a few directories and a bunch of files; 'fpcmade.i386-win32' and 'fpcmake.inc'.

This kind of cripples your RTL set of instructions aswell. :? You may want to correct this before I go posting it. Everything else seemed to be detailed enough to follow. Only exception is the location of a good GNU make utility to use if you choose to NOT use msys, which I opted for myself to reduce the required amount of install packages and things one has to go look for and update later on.

MSX: The reason that the current ARM code is so slow is because that the GBA's normal ARM instructions are a bit slower than the ARM7 THUMB instructions. Don't ask me why, they just are.

Until FPK adds support for ARM7 the compiler will be a bit crippled in this capacity. Hopefully once we manage to establish a nice package for his team to go over and add GBA support into the current compiler, we can look forward to them adding full GBA support into the current roadmap.