Calculations of altimeter error due to temperature deviations from ISA, are, as has been made abundantly clear(?), full of approximations and assumptions. First off, the "rule of thumb" is itself an approximation. Secondly, the appropriate CAA document uses (pace Send Clowns) the words "latest forecast QNH (Regional Pressure Setting) for each hour". Of course, this is a QNH datumed at msl, so ought to be the same as the QFF, except that a QFF is calculated differently, on the assumption that the air mass is isothermal at station temperature all the way down, or up, to be pedantic, to msl, and may be different from an amsl QNH.
For the JAA exams we assume that if we are flying on the Regional Pressure Setting we calculate temperature corrections on the height amsl that we want to achieve. For a 3000ft obstacle, which may or may not be solid, with a 1000ft clearance, we calculate the error over 4000ft. Of course, as Send Clowns points out, that gives a new number for indicated height, so for perfect accuracy we might re-calculate on this figure, and so on ad infinitum. However, there is no point in doing this, for the assumption we made in the Rule of Thumb that ISA deviation was the same at all heights may have already introduced significantly larger errors.
And there is the perfectly valid point that if the first 3000ft is filled with rock, the actual conditions may be sigificantly different from the case of overflying a 3000ft TV mast.
So let us apply the KISS principle. Use the given Rule of Thumb with all its approximations and base your corrections on height amsl if the Regional Pressure Setting is in use, whatever the ground in between.
In real life, the procedure for calculating safety altitude introduces enough factoring to take care of the small errors.
Dick W