Unfortunately not, DEX appears to be a 50% mod whether at level 14, level 99, or anywhere in between.

If I use a 48 fSTR2 cap, 25 fSTR cap, and 50% DEX mod (times alpha like normal), I get a slope of ~0 when I plot the remainder with those values subtracted against STR and DEX. That means that we've accounted for the magnitude of the STR and DEX changes. The fact that there's still noise means that my implementation is not quite correct and there are calculation issues somehow.