i deduce from ur post that u do not yet have a xfmr, only comparing the heat of the 2.5mva one on the 2mva generator and thus think smaller is ok for ur 2.0mva load test. if this is correct, then right out of the gate u can use a 2.0mva xfmr since that is 100% rated.
so question, if above is correct assumption from ur post, seems how much smaller than 2mva can u go?
I would say there are 2 limits: 1) temp rise, & 2) current carrying capacity of wire (gauge) in xfmr.
for 1) I would ask the xfmr mfgr u get quotes from to tell u their thermal time constant. You know then how it will heat - in 1 time constant it will get to 67% of max temp, 3 it wil get to 98%. I think u will find the thermal time constant is about 8 hours so u should be able to downsize quite a bit.
so much so that 2) will be an issue; the smaller xfmr may not be built with heavy enough wire for your current draw at 2mva.... If so I would ask in quote for larger wire size if required.
U can offload both questions to the xfmr mfgr in your request for quote - let them tell u how small u can go! free advice we give when asked for a xfmr quote all the time. also, this offloads the responcibility of it working to them (us)
Also since size & weight is issue for u, I would ask for an autoxfmr to make both smaller (and cheaper) - since a 4000v->400v auto will be 1/10th the kva equiv size of the iso model