Fully Buffered Table vs Internal Table

I had a question regarding performance of a fully buffered table vs an internal table.

In a BW characteristic routine, the current code selects data off a fully buffered table. This process repeats itself for every record which passed through the routine, which could be millions of records each day.

A possible alternative I was thinking would be to load the table into an internal table, which would only happen once, and the selects would occur off the internal table.

Can anyone recommend how much a performance improvement if any, that change would provide? Any help would be great.

Thanks.