The current approach of CSIT code is to attempt to set max frame size on NIC driver to 9200 (usually using hw_interface_set_mtu VPP API call), but any error is ignored and the test continues as if the new max frame size is set correctly. This step is done when setting any VPP interface up, so for example bonding test does that first for the underlying interfaces and then for the bonded virtual interface.

As some NIC drivers do not support setting max frame size, output of the current CSIT code does not tell whether the new value is used (as opposed to the default value), unless it appears in "show hardware" stats telemetry (not all drivers provide max frame size in their output).
for example with SR-IOV, the max frame size is only settable on PF (not on VF), so no devicetest actually sets the value (as multiple device tests are sharing the same PF, interacting only with their reserved VF).

The open question is how should CSIT be handling max frame size (options include not changing the current approach).