I believe it just isn't characterized yet. I did not see any problem report for MII at 2.5V. By the way QorIQ LS1021A Design Checklist, Rev. 3, 12/2016 (AN4878) has table

while both EC2 and EC3 are connected to the LVDD and RMII on the EC3 should be at 3.3V acoording to the data sheet. I also think that requirement that "LVDD and L1VDD must always be the same voltage" also is related to mix of the L1VDD and LVDD signals on some interfaces. Hence the configuration EC1 RGMII at 2.5 V and MII on the EC3 at 3.3V also is possible at some condition with workarround for the managment interface (2 signals) for the RGMII PHY.
However I am not sure. You may try to create a technical case to the nxp support for confirmation.