Looking at the images in your first post...
Pack 12 seems fine and shows that focus is near to the 3438 position. It doesn't have the left hand side of the V but if 9 positions had been used instead of 7 it would have been fine.
Pack 13 seems similar except that there is even more missing from the left hand edge of the V. The last point, at 3246 looks as if it would have been close to focus had more points been used.
Pack 14 is a mess, no idea why but this seems to happen occasionally.
Pack 15 also only seems to be showing the right hand side of the V focus curve with a focus point at 2951. If more than 7 points had been selected this may have given a good focus.
So I would start by endorsing Ken's recommendation that you increase the number of steps. He suggested 9, I think, to start with at least, 11 may be better.
Ken's other suggestions may also help.
One other thing, the focus positions that seem to be found are very variable, the data shows 3438, 3246 and 2951. It seems strange that the focus position isn't consistent from run to run. This will be read from your focuser, is there a reason it could be changing?
Chris