Just relating my experience. I am using a HEX (RB750Gr3) purely as a vlan enabled switch in my production environment.

I had a choice to use bridge vlan (no hardware offload) or switch chip vlan (hardware offload). I was concerned that using bridge vlan would max the cpu (and reduce performance). I didn't have to worry.

In the HEX using bridge vlan (no hw offload), 50 Mbps Rx / 10 Mbps Tx (real world condition) on the uplink/downlink ports (minimal traffic on the other ports) pushed the cpu to a max of 5%.

My conclusion, if you are thinking of using this product (or those with better cpu) as a vlan enabled switch, keep it on the bridge, there is no need to engage the switch chip (although its nice to have the option).

The RB750Gr3 doesn't have a VLAN table in the switch chip anyway, so you can't implement VLANs properly except on the bridge.
50 Mbps is hardly a strenuous test. Try switching 1Gbps on the bridge and see what the CPU figures are...

I did some extensive tests on a RBD52G (a.k.a. hAP ac²) ... it's not really comparable to RB750Gr3 (different architecture, different block diagram), but if one compares published test results it can be expected that RB750Gr3 can do 1Gbps wire-speed switching/bridging (including VLAN filtering) ... at considerable CPU load for sure.