I love the idea of the "Salt" algorithm (where does that word come from anyway?). This would require some doing, but basically:
1) The WCA delegate generates an electronic copy of the scrambles. They would all be of the desired length, as per the current rules.
2) Someone other than the delegate, just before the beginning of the round where the scrambles are to be used, generates a short sequence of turns to applied to the beginning of each scramble in that round.
3) Both the WCA delegate scramble, and the pre-sequence, are typed into a program that is setup to then generate the image of what the puzzle will look like after applying the pre-sequence, then the scramble.
Example:
1) For the 3x3x3 round the WCA delegate generates an electronic copy of the scramble: R B' R' U2 B2 U' F' L' D' B' D U' B2 F R' L2 D2 L B L' F D2 B' F U2
2) Shortly before the round begins someone other than the delegate gives a pre-sequence: L D2 R' F U'
3) (L D2 R' F U') (R B' R' U2 B2 U' F' L' D' B' D U' B2 F R' L2 D2 L B L' F D2 B' F U2) is then typed into a scramble program which generates an image of the cube after being scrambled. This image is either printed at the venue, or the image of the cube is viewed directly off the computer screen.
Now, if we don't want this to result in 30 move scrambles, adding more time to the scramblers' job, then the WCA delegate simply generates a 20 move scramble, and the pre-sequence is set at 5 turns to make the total the same as the current 25.
Pros:
1) The WCA delegate could not prepare the scramble before hand, as doing so would be completely pointless as the pre-sequence would destroy the prepared solve state.
2) Gives an impression that the scrambles are very secure, while at the same time making them actually very secure.
Cons:
1) This would require the use of 2 computers, and also having a printer would be very nice as well. So more equipment is needed than is currently required to run a competition.
2) The "person who is not the WCA delegate" could work together with the delegate. Meaning that the person who generates the pre-sequence could inform the delegate of which pre-sequence is to be used. This could be avoided by having the scrambling program used in step #3 generate a random pre-sequence at the competition, thus avoiding the influence of another person being involved in the first place. To make this more secure the scramblers themselves would be the ones to use the program in step number 3 shortly before they begin to scramble.
3) This would require a program to be written that would allow the staff to type in the pre-sequence and the scramble that would then generate an image of the scramble. This is not necessarily a con, but it would mean more files need to be on hand for a competition to be run. Plus, someone would have to write it.
This seems like a lot of work, but it would at least make things more secure than they are now as long as the 2nd Con issue in my list is addressed properly.
Also, for the record, I am in complete agreement with Mike about generating BLD scrambles. For every competition where I have helped organize I make sure that my co-organizer generates the BLD scrambles, and that I do not see them until after I have done my official solves. I feel that having the WCA delegate see the BLD scrambles before the event would give a distinctly unfair advantage. If I even caught a passing glimpse of the 5x5x5 BLD scrambles beforehand, then I would be able to spot blocks of certain colors and would already have an idea of which re-orientation I might use. The same issue would arise with 4x4x4 BLD. I find it best to avoid this issue by simply having the co-organizer generate those scrambles, and I never see them until after that event is over.