Variability in a completely randomised design

In a completely randomised design, it is possible that one treatment may, by chance, be over-represented in a block whose response, Y, is naturally high, inflating its apparent effect. For example, treatment A would appear to have too high a response mean in the example below.

Block 1
(high Y)
  Block 2
(low Y)
 C   A   A     B   B   A 
 A   C   B     C   C   B 
 B   C   A     B   A   C 

Of course, treatment A could also be under-represented in the first block, resulting in high variability in its estimated effect.

A randomised block design would ensure that treatment A is used for exactly 3 units in each block, so the high response values in Block 1 could not distort its effect relative to the other treatments. Therefore its estimated effect would be less variable (and hence more accurately estimated).