It is hard to tell without seeing the configuration for both your MRTG and Cacti metrics that are being graphed. Presumably you are expecting the two graphs to look identical?
It is possible that the lower values for Inbound being seen in the bottom graph are caused by counter wraparound. To avoid this, you need to ensure that your MRTG does one of the following
- Poll every minute rather than every 5 minutes (the default)
- Use SNMPv2 32bit counters, so that there is no wraparound.
You should also do the obvious checks i.e. verify that both are really graphing the same item, at the same polling frequency, averaged over the same intervals, etc.