So something sends out a broadcast packet and then the layer 3 device splits that packet out to each and every other device on the subnet. Network hardware has to use processing power to do this kind of thing, it's pretty low demand but as a network grows it can get nuts. You'll have all the requests of "Who has [ip]" or "Where is [mac]" etc. A lot of the random traffic then triggers broadcast responses too.

A big issue you have though is if you have any kind of thing doing autodiscovery... like add a printer, angry ip scanner, connect to a networked speaker, broadcast to a screen...etc... it's gonna do some combination of a broadcast, or checking one by one on a specific port (usually starts low and goes high but i'm sure some moron starts at the end of the subnet and counts down.) There's tons of little things that do this and some of them probably do it without us even thinking about it in the background. This might take certain tasks like network share discovery take an inordinate amount of time.

In businesses we usually have fairly high throughput network devices but our home devices less so. My gateway/firewall is layer 3, so is my core switch. I don't think I have layer 3 configured on said core switch right now. My firewall ends up being my router and it has a NAT throughput of like 950mbit. My isp is gig fiber, so I already can't utilize all of that. All the random shit going on in my network that needs to route also eats into that throughput because the firewall has limited processing power and memory. Then you start factoring in things like vpn (which my firewall hosts) which will drag down my throughput a shitload when in use and you start seeing where efficiencies can be more important.

Thankfully though, I don't do much of anything important and my network is already overkill for anything I do. I do have options to offload the layer 3 traffic through the switch if it ever comes to it and simply have the gateway/firewall behave as a gateway for internet only.