The “hold" you are using inside of xmult is not the same as the global “hold”.
So, presumably you set the global hold to be 100,000 trees.
After getting 14 trees from various parts of tree-space, you then applied tbr branch swapping, and when the number of equally parsimonious binary trees hit 100,000 you ran out of room for more trees.
All it means is that there are more than 100,000 equally parsimonious trees at 2663 steps.
But this is not a big deal since you weren’t likely to get ro trees that were more parsimonious than the 14 at 2663, you were just going to get more of them.
The appropriate thing at this stage is to do a consensus with nelsen;
That will tell you the extent to which the data are decisive about finding various clades.
It probably will not be much different from a consensus of the 14 you had in the first place.
I don’t know of any situation in which looking for more than 100,000 equally parsimonious trees has been a fruitful/meaningful thing to do.