Some suggestions:
Once a player is spotted, maybe ignore any checks on that player against players that spotted him for, say, 1 second (customisable for testing?). This may help prevent any 'flickering' of players and improve performance, especially on ramps. It's unlikely a wallhacker could use that time to much of an advantage. I'm guessing there will be points on some maps where checking rectangles on each frame will cause a player to flicker and disappear - it's unlikely enough points can be always checked to ensure a player can be seen. Disappearing/flickering players are bad and lead players to have little confidence in the server.
Ramps are probably going to have the most issues where only a tiny gap is peaked and a very limited portion of an enemy will ever be seen, so maybe increase the points checked when players are at different heights or allow server owners to determine number of points to be checked.
As player numbers reduce, use more checking points for models.
In smaller servers allow heavier checking of model positions Maybe allow the number of extra points to be customisable? And / or maybe code it so more points than just the corners to be checked as player numbers reduce?
Maybe more like:
[IMG]http://img153.**************/img153/9165/wall02s.th.jpg[/IMG]
I'll be honest here and say my primary reason I want this is for mixes/scrims where client-side anti-cheats aren't used. However, I do help run and admin one of the most popular UK public servers and totally appreciate the help it gives public admins.
I can see that checking model points will never be perfect unless someone writes and exceptionally efficient hit box detection routine and removes any players that can not be seen from being updated in the netcode (go on HPE/VALVE you know you want to...). However, the more points that can be checked the better.
Maybe allow another toggle so that only players from one team are checked to further reduce the number of checks necessary on each frame. This would be useful on large servers where a wallhacker is suspected on one team.
And finally something strange spotted while testing, on Dust2 for some reason it allows TS to see through the double doors whilst CT can't (using r_drawothermodels 2).