The sad thing (but that's true for all but the highest-end devices) is that any advanced stuff (VLAN filtering included) is done in software and then it doesn't really matter which switch chip is used, it'll be the interconnects between switch chip and device's CPU that will become bottleneck.
True but there's an interesting observation to be made here:
Take a hAP ac unit which is a 720MHz single core MIPS device and the RB4011iGS+5HacQ2HnD-IN which boasts a quad core 1.4GHz Cortex A15.
If you have a few VLANs in place, i.e. one for hosts (PCs), one for each CPE (internet gateways), one for VMs and one for servers (NAS units, Hypervisors, UPS devices, Raspberry Pis, etc) - then you most likely would want to have some mangle rules running 24/7, thus requiring you to keep the "Use IP Firewall" and "Use IP Firewall for VLAN" bridge settings on.
Let's say that we want to move a big file from our NAS, such as a Linux Distribution iso, to one of the hosts so we can create a bootable USB installation media.
On the hAP ac the processor will instantly shoot to 100% and our transfer speed will not be able to surpass 110MB/s.
On the RB4011 the processor will instantly shoot to 25% (100% on one core) and our transfer speed will still not be able to surpass 110MB/s. Even though that one core offers double the frequency compared to the older model's CPU.
I wouldn't personally believe (as the transfer is single threaded) that it's because that the MIPS processor is twice as efficient - simply because the RB4011's CPU can greatly outperform the hAP ac in intensive tasks such as packet marking, and with about 30 times less CPU usage too. (Test was conducted with 4 packet marking rules and 9 queue trees)
The hAP ac can achieve 125MB/s transfer speed with tolerable or minimal CPU load, if the "Use IP Firewall" option is disabled.. but that would render the whole mangle table useless, as every interface is basically a bridge slave.
So my closing statement is the following:
I'm okay with doing things in software, as it makes them a tad easier to configure and nowadays devices have enough power that indeed (performance-wise) makes it no different than what the hardware could've instead offered. But even though the hardware is becoming more powerful, in many orders of magnitude, why do we still have to pay such a ridiculous performance penalty for basic things?
And to anyone that's about to go ahead and do it, please don't kid yourself by saying that: a 230EUR device shouldn't be used for VLANs, mangle and routing combined.