Five minutes, even outside of low latency trading or financial services, is an eternity, from an operations perspective. Five minutes is really only ever going to be good for a capacity planning or an indicator type function. As soon as you're interested in operational response and getting your tier three ops team onto problems as soon as possible, that's when you need to bring down that polling granularity to one minute.
Now, there's obviously an observer effect issue here. You don't want to bring polling to such a level that you actually effect the operation of the inline device. There's always a trade-off, and there's always a trade-off depending which mid value you're going after, as well. If I've got a five minute average count, why do we go and poll that every one second? It makes no sense. It's not really going to give me any meaningful information. The ability to drop down to those levels and to do it on a large scale, and to do it for specific elements of the network, is very important.