Personally I always go for the 'Any plan has a chance of succeeding' style. I would have let him try if he wanted to, but told him in no uncertain terms that it would be a difficult series of checks to pull it off and that failure could result in injury or even possibly death.

If he still wanted to proceed I would have probably started off with a moderately difficult acrobatics check to stay on the ballista (failure = fall damage and the bolt goes on without you), a reflex save to avoid the acid when the bolt when off (failure = take the damage), and some kind of modified grapple check (maybe grapple vs. a set DC? Or a climb check to 'catch' your fall) to grab onto the dragon (failure = fall damage but you're now adjacent to the dragon). Net result on all successes would be some kind of atk/dmg bonus from being up there, something enough to make it worth it, but not enough to completely shift the battle (if this was 4E I would do something like +1 atk and you can add your weapon critical damage to non-crits. In 3.5 probably +1 atk and reroll the lowest 1-2 damage dice, something nice and a reward for the hard work, but not something that auto-wins the battle).

I've found that allowing players to run with their crazy plans but keeping realism in there somewhere (while allowing it to be thrown out the window by a good set of rolls or a high modifier) gives the best balance of chances for epic moments and more mundane battles (that let the real epic moments really shine, can't have any highs if you don't have a baseline).