The nightly rollup job takes longer than expected, and you might observe a global slowdown of the core's jobs.
The disk activity where the repository is located is at nearly 100% most of the time.
If you expand the rollup job , and isolate the machines that takes significantly longer to rollup than others with a similar amount of protected data, you should be able to define if those are machines with an intense memory usage type of workload, such as Exchange in general.
Once these machines are identified, browse to the recovery points list of one of them, then detail the recovery point to show how is the recovery point's size is spanned across the different partitions.
Should you notice the C: with larger than expected incrementals, and roughly more or less the same in size on all recovery points, this would indicate that the pagefile is generating significant block changes, that are backed up.
The problem lies in the fact that the changes are constant in the pagefile, therefore generating random blocks. The compression and deduplication against such type of randomly generated blocks is by nature very low.
The direct consequence is that the rollup of such data is taking much longer as all of this data is effectively written to the repository. When the rollup consolidate the recovery point chain, it does work on this amount of random blocks , and then needs to clean the left over once the rollup has completed.
The whole process is extended due to the recurrence of similar behaviors among the protected agents.
In order to take more efficient backups, it is strongly advised to move the page file of these machines as soon as you can schedule a reboot of the machine.
Moving the page file must be done to a disk that is excluded from protection.
Doing so ensure that only needed data is going to be processed, consolidated and checked, in a much faster way.
You should observe as a first result of these changes, that the next incremental recovery points for these agents have now very small incrementals for the C: drive (ex : 2GB before , 130 Mb after).
Should this be the case, you should see over time , as old recovery points including the page file data fall out of the retention policy, a decrease in the time needed to complete the nightly rollups, and an overall improvement regarding the jobs speed and the disk usage.