That's pretty strange. You could try to find out if it's as fast as normal RAM (or TCM, but that seems unlikely)
I have to correct myself. With a bit more careful testing, it turned out that this memory area is actually 0x2000 bytes long and repeating between 0xbfe00000...0xbfefffff. That makes me think it might be a TCM of the other core.
Using the same test loop, I got the following results for regular memory, btcm and this area:
MEM c(1) 28320 R, 38085 W Kb/s
MEM uc(1) 1953 R, 38085 W Kb/s
MEM c(4) 64453 R, 151367 W Kb/s
MEM uc(4) 10742 R, 41992 W Kb/s
MEM c(16) 88867 R, 214843 W Kb/s
MEM uc(16) 42968 R, 191406 W Kb/s
BTCM (1) 37109 R, 0 W Kb/s
BTCM (4) 153320 R, 0 W Kb/s
BTCM (16) 615234 R, 0 W Kb/s
BFE1 (1) 12695 R, 38085 W Kb/s
BFE1 (4) 50781 R, 153320 W Kb/s
BFE1 (16) 180664 R, 611328 W Kb/s(Some or all of these results can be buggy. 1, 4, 16 are the access units - byte, word, 4 word)
One thing I've wondered but never got around to experimenting with is how much (if any) of the BTCM is used on cams that don't copy dryos code there. Even on the ones that do copy, there might be a fair bit available (sx710 copies ~30kb)
The sx280 appears to use some at its start and some less at its end, but most of it is just noise. I just dumped this area several times (while doing something on the cam and also after off/on cycles), and compared parts of the dumps visually.
edit:
The m10 also has this 0x2000 byte area repeating between 0xbfe00000...0xbfe0ffff and 0xbfe20000...0xbfefffff