Thanks, it would have minimal impact. Don't forget that a sustained read across all drives (the scenario) will flood the Thunderbolt bus, so this scenario of 2 drives going faster would have no real world impact, as with sequential reads on a volume, all drives would be requested to deliver data in order. which it will do at 2800MB/s.

At least we know this is a hardware design, not a faulty enclosure or drives.

This is a part of the hardware design apparently. I tested this here and see what you are referencing, so it is not a problem with your blades.

It appears the first two drives (PCI wise) get priority for read requests. (similar to my earlier description, but I am sure I did not describe it accurately from an engineering perspective)

There seems to be no difference on write requests when I tested.

This is a complex design to get 8 blades into one enclosure. You are likely measuring an artifact of that. It has no impact in the "real world", so lets ignore this.

I can't explain, but will discuss this.

It's clear that fewer disks results in higher performance, but there are themes and variations: disks 1 & 2 are always green no matter how many drives are verifying, and 1-6 (six disks) gets better results than 3-8 (also six disks!).

This is how it was explained to me and something we have seen since Firewire.
If you certify 6 disks only, I believe you will see the same behavior. (3-8)

I'm not exactly sure how either of those follow. I always get 100% performance out of disk1 and disk2 in particular no matter how many disks are active, and disks 3-8 drop down to 50% performance when the enclosure is saturated. So I couldn't say it's a disk1 > disk2 thing, or that the difference is merely "some" latency.

There is actually an answer. It has to do with how IO's are queued with MacOS. If two complete at the same time, disk1 gets priority over disk2, so this reduces some latency on the first two drives.

Sorry, you did say that. I can inquire about this, there should be no difference between different PCI lanes.

No apology needed! My hypothesis is not that there are consistently faster lanes (this seems disproven by testing), but that there can be bus congestion and prioritization when whole device is jamming. I'll be interested to hear whatever explains the results I'm seeing, but it doesn't make the device unusable or problematic.