There may be a work around for this issue that is already built into the code.

Can you try the following steps:
(1) Create a new 4D file that only contains your 20 ROIs as 20 "voxels";
(2) Estimate the SPM model using the new data;
(3) Add the following to your parameter structure:
P.FSFAST=1;
P.FSFAST_source=[SPM.mat filename and path to original data];
(4) Run the PPPI commands

This will extract the seed values from the original dataset and then compute the PPI models on the reduced dataset.

NOTE: I have not tested the order of operations to see if averaging --> estimating is different than estimating --> averaging are different.

Let me know if you have any questions.