Neither image is "thresholded." If you load and view either the con or spmT .nii files, you will see that there is a value at each voxel.

The difference between the images is that the con image gives you the p-value at each voxel, while the spmT image gives you the t-statistic at each voxel.  The functionality is the same as in regular null hypothesis significance testing.  The t-statistic accounts for the difference in observed Beta relative to standard error of the Betas.  The p-value tells you the statistical significance at each voxel; therefore, it accounts for the t-statistic at each voxel as well as the degrees of freedom of the data.

P.S. when you do a second-level analysis, you should always use the con files, not the spmT's.