GNU awk weirdness with array indexing in loop

I was drafting a solution to a question earlier (this one actually), a simplified version:

$ echo {a..g} | tr ' ' '\n' |  # create some data
  awk  -v s=e '{               # search string as an argument
      for(i=1;i<=2;i++)        # keep 2 last records in a
          a[i]=a[i+1]          # this is the problematic part later
      a[3]=$1                  # hash the current record into a
  }
  $1==s {                      # once there is a match
      for(i=1;i<=3;i++)        # output from the hash
          print a[i]
  }'

Output:

c
d
e

Being me I had to try and make it a few bytes shorter (remove i++ from the for and a[i+1]-> a[++i]:

$ echo {a..g} | tr ' ' '\n' | 
gawk -v s=e '{
    # i=1; while(i<=2)  # fails with while also
    for(i=1;i<=2;)      # i++ moved
        a[i]=a[++i]     # ... here
    a[3]=$1
}
$1==s {
    for(i=1;i<=3;i++)
        print a[i]
}'

But the output failed on GNU awk:

    

e

After a while of wondering I switched to mawk and it worked. It also worked on Busybox awk but failed on awk-20121220. Any idea what causes the behaviour? Also, if you guys have a more descriptive idea for the title, feel free to change it.