The program doesn't "know" that you just masked the data with your 3dcalc expression. The author (me) assumed the worst, that you changed the values in some arbitrary way, so that they are no longer from well-known statistical distributions -- for example, if you took the cube root of a t-statistic, it would no longer be useful.
You can re-attach the codes that mark various sub-bricks as coming from known statistical distributions using the 3drefit program's "-substatpar" option, as in
3drefit -substatpar 2 fitt 49.0 dataset+orig.HEAD
which would tell the program to mark sub-brick #2 (recall that counting starts at #0) as a t-statistic with 49.0 degrees of freedom. See the output of 3drefit -help for more details:
http://afni.nimh.nih.gov/pub/dist/doc/program_help/3drefit.html