They had something like 58000 atoms in their lattice and managed to form 17000 qubits from atom pairs.
Pretty impressive already yeah, I'm sure either their team or another research team will try to improve the qubit yield. Increasing the atom count could be harder though, I'm not an expert but at some point you run into limits from the size of your lab/cooler/trap.
And yeah 99.9% fidelity is not anything groundbreaking, but it's pretty surprising for a "first try" on a new gate implementation. In contrast superconducting systems have been worked on for years and years with ridiculous amounts of investment and also only get 3 nines of fidelity, and IonQ just recently demonstrated 4 nines fidelity, but only after working on their system for many years too.
It will be very challenging to implement their system off this paper as there are many key experimental details missing. I won't get into my hatred of this but open methodology is sorely needed in QC and it really hampers speed of development across the whole sector. As you say you will run into limitations eventually but again hard to judge where those limitations are.