The CPU utilisation depends on the amount of traffic through the box. Matching each packet to the correct flow takes a tiny amount of CPU - so the more traffic that flows, the more CPU that's needed.
Also, exporting those flows from the box takes some CPU too. You don't have to export the flows, but then there's really not much point in gather the information in the first place - unless you're going to run a feature such as netflow top talkers right on the box itself.
As for your second question: there are two timers for flows:
The active time ("ip flow-cache tiemout active") controls the longest amount of time a flow can be active for before it's exported. This ensures that the collector gets to know about long-lived flows.
The inactive time ("ip flow-cache tiemout inactive") controls how quickly flows are exported after the last packet is seen. If your traffic is quite bursty then you might want to set this a little higher - though setting it too high means your netflow cache will be holding on to a lot of old data.
Finally, note you can also adjust the size of the netflow cache itself with the "ip flow-cache entries" command.
