If you are running this analysis single-threaded, it is possible that it would take that long. Keep in mind that the early time estimate is at least 30-40% higher than it should be. It will stabilize after 20-30 thousand generations, and from that point it likely will be accurate. Separately, if this is a single protein that you are analyzing 2.5 million generations may be an overkill.

For a multi-threaded version (compiled using OpenMPI) you can potentially cut down the time by 8x - presumably you are doing 2 runs with 4 chains, so those 8 can go in parallel.

You may want to add couple of extra lines:

lset rates=gamma ngammacat=4;
prset aamodelpr=fixed(MtMam) shapepr=fixed(0.123);
mcmcp filename=name_your_output_here;
mcmcp nruns=2 ngen=2500000  nchains=4 printfreq=1000 samplefreq=100;
mcmcp relburnin=yes burninfrac=0.25 mcmcdiagn=yes diagnfreq=1000;

Afer mcmc, to to get the summary:

sump relburnin=yes burnin=0.25;
sumt relburnin=yes burnin=0.25;