First off, if you have not upgraded to either the latest core version of 7.2, 7.3, or 8.0, I recommend you do so as there have been updates that optimize the acceleration process.
Secondly, you should check to see if your datamodels have high cardinality fields, i.e fields with a high number of unique events. This leads to additional memory consumption on the Search Peers.
One way I like to do this is by using:
|datamodel <DM_name> acceleration_search
| stats estdc(*)
Running this for a short 15-minute window in the past will tell you if any of the fields in the DM have high cardinality. If possible, either filter out some of the values in these fields or remove them for the DM definition.