I have 3 MikroTik switches: 2x CRS317 + CRS326, 2 MikroTik routers: CCR1009 and hAP ac², 1 load balancer. Hardware resides in 2 racks and is connected in following manner:
| symbol is border between racks. LB has 4 physical gigabit interfaces and due to lack of space in second rack it had to be placed in first one. Because of that I need LB ports to be "directly" exposed in. second rack in CRS326, especially two interfaces that are used for lab setup and their config changes really often. Obviously I don't want to reconfigure VLANs on all switches along path every time it happens so I want to directly carry traffic along with all VLANs defined on LB to 4 specified ports in CRS326.(two of which are connected to 2 routers so I want it to behave like LB would be connected physically to routers)

According to my understanding of topic this can be done using Q-in-Q. Unfortunately I noticed that it requires changing ether-type on bridge, which, due to limit of 1 HW accelerated bridge, seems to collide with standard VLANs on switches.

However then I noticed VLAN stacking config which seems to use standard 0x8100 ether-type.

So what's difference between tag stacking (0x8100) and Q-in-Q (0x88a8)? I mean yeah I understand it's physically encoded in different way and it's separate standard and that some other switches may only support conventional Q-in-Q with SVID but in this particular scenario (full mikrotik CRS3xx) is there any functional difference between Q-in-Q and CVID stacking (apart from that one method seems to be incompatible with normal VLANs on the same switch)? Are both methods hardware accelerated on CRS3xx?

Yes, both methods are hardware accelerated on CRS3xx and they work.

Question regarding functional differences remains open.