Question: Is it a good practice to use "--scan-time 1" in cgminer?

I'm trying to solo-mine FTC (scrypt, 2.5 min/block) with GPUs. From what I can found, the usual recommended practice is to add stratum or longpoll pools to the list of pools, so cgminer will know when there is a new block.

However, from my own experiment, I found that existing FTC pools are slow to send out new block notifications. The following is a screenshot of cgminer running on 2 different machines.

http://i.imgur.com/AphbAkv.png

Machine on the left:
pools = [feathercoind, (stratum) feathercoin.is-a-geek.com, (longpoll) featherpool.com]
scan-time = 60

Machine on the right:
pools = [feathercoind]
scan-time = 1

Using "--scan-time 1" with a local feathercoind gives me new blocks detections 0-30 seconds faster than the stratum/longpoll connections (as seen above).

However, the statistics for the machine on the right show 60 new blocks (NB), 177 discarded work items (DW), and 1 rejected block (R). Having "DW > NB" and "R > 0" makes me feel that it might I may be doing something wrong. Naively I think that cgminer discards work only when a new block forces it to do so.

If anyone knows whether it's OK to use "--scan-time 1", and whether it's fine to have "DW > NB" and "R > 0", please let me know. Thanks!