predict using a CompactRegressionGP object is much slower than using the full GP object

I have trained GP model for regression using fitrgp.

When using the model to make predictions on new data I use predict:

>> tic, [mu, sd]=predict(Mdl_full,Xtest);toc

Elapsed time is 0.070162 seconds.

Mdl_compact = compact(Mdl_full);

>> tic, [mu, sd]=predict(Mdl_compact,Xtest);toc

Elapsed time is 4.185938 seconds.

Why using the compact model is much slower than using the full model? Any way to make it faster?

Note that the compact model takes about 2000 times less memory, hence it is better to use especially as a member of an ensemble