The grid pattern/vertical striping is limited to revision 3-6 because on this revision there is an area called "NTSC only" and the RGB lines are split before they go into the RGB encoder and sent to this area:
http://img215.imageshack.us/img215/9176/aes36be1.jpg
The problem is that these traces run under the clock signal for the encoder and cause some weird interference that shows up as the infamous vertical striping. If these traces, which are totally useless anyway, are cut/lifted from the PCB the striping effect disappears and you get a flawless RGB picture. Earlier revisions don't have these split RGB traces and consequently no striping/checkerboard effect whatsoever:
http://img85.imageshack.us/img85/2975/aes35yl1.jpg
That gamesx article fails to explain the phenomenon and claims that revision 3-5 (the '126000 serial') has a partial striping effect ('checkerboard'), i.e. better RGB than 3-6 but worse than the preceding revisions, which is simply false.
Revisions 3-5 and 3-6 have a brighter RGB output as a result of different values for coupling caps and resistors but this obviously doesn't "dramatically decrease the RGB quality". In fact the new values (470uF/75ohm vs. 100uF/68ohm) are what Sony recommends. In the end the difference between the darker RGB of the earlier revisions (NEO-AES, AES3-3, AES3-4) and the brighter RGB of the later revisions (AES3-5, AES3-6) is trivial and can be compared to a supergun setup with adjustable RGB pots where some will prefer the picture a little darker while others a little brighter.