can you set the Normal check interval for service checks for the Check_MK HW/SW Inventory for this device to e.g. 10 minutes and confirm that you see spikes every 10 minutes please ?
I am working on a similar issue for a customer:
He also get s impossible Interface spikes but in this case the device was not reachable for about 15min for the monitoring system. I was not able to reproduce this behaviour so far.
Can you also provide one rrd file of an interface showing impossible spikes ?
Because what I observed in the mentioned case is that not only one data point was impossibly high but 3 in a row.