I calibrate our cameras in-house using the big dot wall. I ran it both ways - with my camera info and letting Agisoft pick the interior orientation values. Entering the camera calibration file into an Agisoft project and it did not improve the error results. In fact, it was less accurate by 0.03 m (in the Error M and Error Pix columns). I didn't do an accuracy assessment in ArcGIS though. Letting Agisoft pick the interior orientation info was slightly better in the end.
The only time I'd use that camera calibration info is if I created my mosaic in EnsoMosaic which needs it.
Someone asked me why I thought my calibrated camera info would improve results (I don't know it just seems like knowing exact data would be ideal, no?), but I was surprised.