Well, it depends on the accuracy of the GPS info of course. If the GPS data is not resulting from a D-GPS solution, the data are just good enough for image pair selection and global georeferencing of the model. However, it can not be used for optimizing the camera parameters since the accuracy is much too low.
Concerning the amount of feature points: often, lower amounts give much better results. Half a year ago, I rigorously tested the accuracy of camera parameters and image alignment of a 60 image dataset. There was no single trend to be found: sometimes, accuracy and alignment were high using 20 000 points, sometimes the alignment completely failed. Often 40 000 points delivered worse results compared to 20 000 points, but 35 000 points was again better. So, no single conclusion could be drawn at that point. I did not execute these tests with the newly implemented algorithms, but I suppose results will be largely the same, since the SfM step has multiple mathematical solutions. My preferred workflow is to align everything using 20 000 points and afterwards filter the points based on reprojection error. After deleting the points with a reprojection error higher than 0.5, I optimize the point cloud and filter again, until the end results delivers me a maximum reprojection error of 0.5.
Cheers,
Geert