If you really want to salvage this mistuned-TR run, it would be better to analyze it correctly with the TR as-gathered to get the beta weights, and then combine those with the weights from the other runs. With 3 'good' runs and 1 'bad' run, you would combine betas as 0.75*goodbeta+0.25*badbeta (e.g., in 3dcalc). This presumes you are taking the betas to the group level. If you are interested in combining the t-statistics, or something like that, at the individual level (e.g., to make a retinotopic map for some hideous neuroscience purpose), that would be possible, too, but would require a little more thought. But at the moment, I've got to eat a chocolate in honor of Lincoln's birthday.