Couple of remarks:
1. FPC
can actually support 10 bits extended on win64, but it is non-standard (in the sense of non-abi compliant) and requires a recompile of the compiler and rtl. Not recommended unless you really need it. Florian explained somewhere how to do that. (can't find it soon enough, but the info is there)
2. The issue at hand is the same for (48 bit) real, which is aliased to double, but has a support unit to be able to read/write and to some extend calculate with 48 bit real values. (unit real48, which is in reality the real real and not a double
)