It seems that block validation time is almost linear in block size, i.e. t = O(blockSize). If we reduce blocktime, blocks will become smaller and validation time will decrease correspondingly. In other words 10 times more frequent blocks are 10 times smaller and require about 10 times less time to validate. Am I missing something?
No, that's pretty much correct, I think. It's also not hard to show that a miner's orphan rate can be written approximately as
1 - e^(-t/T)
where T is the block period, and t is time required for the other miners to download and verify the block*. t scales with block size, as you noted, so clearly you can scale block size at the same rate as the block period, and the orphan rate is left invariant.
Thus, a system with a low transaction rate can safely get away with a smaller block period, but keep in mind it will at some point have to raise the block period to accommodate a larger transaction rate, in order to avoid wasting too much hashing power on orphans, or screwing up consensus forming.
Much better IMHO to just focus on ways of making 0-conf transactions less risky and work on building a network of payment channels* than to try to decrease the block period. It's not like any block period tweaking is going to improve brick and mortar transactions anyway, which need to be done in seconds, not minutes. Although there definitely does appear to be a dumb marketing gimmick aspect to it, which is unfortunate.
* More precisely, t is the fractional hashing power-weighted sum of the times required for the other miners to download and verify the block.
*
https://bitcointalk.org/index.php?topic=244656.0