In reply to GrahamD:
It would be worse if it worked in the opposite direction (as an upgrade would then immediatly result in pressure for a downgrade
but I suspect on routes with very diverse opinions, the average impression for the grade of a route might change by close to half a grade just from the change of grade category following an upgrade (with the same people voting for the box that 'suits' best).
An example with not exactly a huge spread: 100 votes on a currently categorised tough VS that should be easy HVS. 2 think E1, 7 think tough HVS, 17 mid HVS, 30 easy HVS, 25 tough VS, 13 mid VS, 5 easy VS, 1 HS. Clearly averaging as just HVS.
In the 'best suited' voting categories for VS you get 56 'HVS' votes with a scatter of votes through VS to HS; this 'looks' very close to the border but the editor decides to upgrade. The new HVS votes are now: 2 E1, 7 tough HVS, 17 mid HVS 30 easy HVS and 44 VS; the average now looks firmly in the HVS band, higher than the true average.
Now on top of this effect firstly add the distortion of inexperienced climbers now chasing the 'easy for the grade' tick (who will almost inevitably tend to overgrade on average and who may have avoided the route with its old grade and a tough reputation). Secondly remove some votes of experienced climbers who can't be bothered to vote again. Before you know it a borderline route with a wide voting range becomes mid-grade.