Patrick,
The scenario I gave was for High volume data and not running this merge command after every update resulted memory growing exponentially. To solve it, we used chunks logic and after every chuck of inserts we ran the Delta merge so that Delta merge memory don't grow exponentially.
We needed to load 6 year of data.
In our proc we had start time and end time as in put parameters and with in store proc called the exec('commit') option and followed by merge option.
This way we made sure that delta merge was never growing exponentially but growing at limited rate.
Hope this explains.
Regards
PK