First check the involved clients if they still have their IP addresses and routes set, as this would be the most likely cause of problems (not renewing the DHCP lease correctly, if that actually is the case, testing more current development snapshots could be a way out).
Other than DHCP woes, the problems you describe generally shouldn't happen, as the router's switch should more or less be configured to pass everything on undisturbed between the LAN ports, so even if the router's OS would get stuck, the switch should remain in its previously configured state... What could result in very similar behaviour would be either loops between LAN segments or faulty devices 'reflecting' some/ most of the packets going by (e.g. years ago I had a switch exposing that behaviour whenever it lost power/ was switched off), which could effectively drown the network with useless packets. Therefore it would make sense to reduce your network down to just two clients connected directly to your LAN ports, check if the problems are gone and then carefully restoring the whole networking infrastructure piece by piece, to eventually pinpoint a potentially broken participant (could be a client's network card, some rogue dumb switches or APs or even some semi-broken (patch-) cable). If you're familiar with network sniffer like wireshark, this might also help to cut down your experimenting.
Keeping in mind that your device is relatively new to market and remembering the recent thread about NAT leakage on TL-WR1043ND v4 could also imply a more general problem with initializing the particular switch chipset used in your router. Also give the vlan setup a quick sanity check.