From the tests I carried out assuming an optimal network condition, where all the ports are UP, the traffic passes through the bonding interfaces, without therefore asking the peer port. The traffic on the peer port passes when the client must necessarily do so from the peer port to reach the destination, for example:
Switch 1:
Bonding 1> Client 1> SFP1
Bonding 2> Client 2> SFP2
Switch 2:
Bonding 1> Client 1 SFP1
Bonding 2> Client 2> SFP2
The two switches can be seen from the peer port, for example SFP24
If you interrupt the SFP1 / Client1 / Switch 1 and the SFP2 / Client2 / Switch2 you will see that the traffic passes from the SFP24 or the peer port ..
This is not the behavior I have seen with my MLAG configuration.
2x CRS354-48G peer'd via QSFP 1-1 (01/02)
MLAG 2x2-4Port total LACP to adjacent switch (Procurve)
A third CRS354-48G (03) is connected to switch 354-01 via single QSFP 2-1
Video server connected to 354-03 while cameras are fed from Procurve.
I see Bulk of data streams going to 354-02, traversing the PEER to get to SW-01 to then traverse the standard QSFP link to 354-03.
Even though the most direct path is MLAG to 354-01 > QSFP > 354-03.
MLAG_example.png
So while in your example two clients MLAG between the two switches directly may work as expected, it's not always the case and PEER link capacity does need to be considered.
You do not have the required permissions to view the files attached to this post.