You might be correct that varience has a near-zero effect for lower-confirmation transactions, but still remember that shorter block times give less varience.
Shorter block times means the attacker can make more attempts.
So someone with 49% has a near-zero, close enough to zero, chance. The actual probability of getting 7 blocks in a row is less than 0.6% for a 49% holder. The actual probability is smaller, because you aren't rolling an infinite amount of dies.
Think it can't happen? I had to search long and hard to find a real example. I had to go all the way back to...today.
Bitcoin network right now: 12.224 Thash/s
Deepbit network right now: 5282 Gh/s
How many block can deepbit get in a row with 43% hashpower? Blocks 143334 - 143341 are deepbit blocks. Deepbit just solved 8 blocks in a row with only 43% hashpower. I'm sure if someone analyzed the blockchain we'd find even longer runs of deepbit-only blocks.
Variance = someone with <51% can successfully attack the blockchain.
Faster block target speeds = more chances to attack = weaker security.
And? This means deepbit must be CONSTANTLY trying to double-spend. It would be wasting a lot of money doing that! Faster block target speeds = less varience, and more oppertunities also means more money lost for the attacker. It's a waste of time faking blocks - better just fork the blockchain (which is a problem for faster-time blocks, since more natural forks for the attacker to "cancer node" exploit are possible.)