A few days ago, I ran df -h for no reasons as I usually did, the output as follows:
$ df -h
Filesystem      Size  Used Avail Use% Mounted on
rootfs           72G   49G   20G  72% /
[...]
To my amaze or not to, the part of my brain which manages mathematics finally told me that 72G != 49G + 20G. I have been using Linux full-time for more than five years and probably have run df for a thousand times if not less. Finding the inconsistency is not really what confused me but why it took me so long to see the numbers.

I found the answer after consulted Google using columns of df, ie. "df size used avail". Yes, that's how I found the answer at my first try. I am still a master of Google.

The discrepancy comes from Reserved Blocks. From what I read, Extended File System (ext) since ext2, has such feature and the default reserved blocks are 5% of total blocks, marked by mke2fs.
-m reserved-blocks-percentage
  Specify the percentage of the filesystem blocks reserved for the
  super-user.  This avoids fragmentation, and allows root-owned daemons,
  such as syslogd(8), to continue to function correctly after non-privileged
  processes are prevented from writing to the filesystem.  The default
  percentage is 5%.
To see how many blocks are reserved, you can use tune2fs:
$ sudo tune2fs -l /dev/sda3 | grep -i block
Block count:              19037025
Reserved block count:     951851
Free blocks:              6019654
[...]
Block size:               4096
[...]
In my case, it's 19,037,025 * 0.05 = 951,851.25. Round to 951,851 blocks as you see in the output above. To do the math correctly, use 1K-blocks for calculations:
$ df | head -2
Filesystem     1K-blocks     Used Available Use% Mounted on
rootfs          74953252 50727968  20417880  72% /
The reserved size is 951,851 * 4 (Block size, 4096 = 4*1K-blocks) + 507,279,68 + 204,178,880 =  74,953,252. Mystery solved!

5% (3.6G) is really a lot, even for my tiny harddrive. Imagine a 1TB harddrive, that's 50G, almost as big as my harddrive. There are two ways to set the reserved amount of blocks using tune2fs, one by percentage, another by number of blocks.

For percentage using -m, the following code set to 1%, you can assign a float number.
$ sudo tune2fs -m 1 /dev/sda3
tune2fs 1.42 (29-Nov-2011)
Setting reserved blocks percentage to 1% (190370 blocks)
Or, the specific amount of block using -r, the following code set to around 500MB in 4K block size.
$ sudo tune2fs -r 129008 /dev/sda3
tune2fs 1.42 (29-Nov-2011)
Setting reserved blocks count to 129008
When creating an ext2/ext3/ext4 filesystem using mke2fs/mkfs.ext2/mkfs.ext3/mkfs.ext4, you can only use -m to specify a reversed percentage.