The issue is that a bonus, in this circumstance, isn't necessarily good - you aren't trying to roll high. Rather, you're trying to hit a specific number. Which means that rolling higher MIGHT be good. To say that you're adding the bonus beforehand means that you could be potentially adding a number that makes you fail (if, for example, you already successfully rolled a 1,5, with no Experimental bonus). Which means that your good luck...gave you bad luck? Which contradicts the point of the virtue.
Hence my "you have to make that judgement after you roll, or otherwise it may be bad luck" argument.
Note that I've also pointed out that the larger your Experimental bonus is, the more skill-based the roll becomes, and thus (arguably) the less luck plays a part, as discussed more below.
Actually, someone with an Experimental bonus of +3 will always add +3 to their roll. At that point, you can THEN choose to either add up to an additional +/- 3 to the final result, after you roll. (This is the official errata, rather than my partially-remembered version.) So, the total swing will end up being from 0 to +6. So, with 3 points of (non-variable, mandatory) Luck, the swing would be +3 to +9. EDIT - which puts us into the awkward role of trying to explain the following probability curve.
Without Luck = valid rolls are 4,5,6,7,8,9, 1/2, 1/3, 1/4, 1/5, or 1/1/2 = 64.1%
+1 (mandatory) Luck = valid rolls are 3,4,5,6,7,8,9, 1/2, 1/3, 1/4, or 1/1/2 = 73.1%
+2 (mandatory) Luck = Valid rolls are 2,3,4,5,6,7,8, 1/2, 1/3, 1/4, or 1/1/2 = 73.1%
+3 (mandatory) Luck = Valid rolls are 1, 2,3,4,5,6,7, 1/2, 1/3, = 72.0%
This is the problem with trying to hit a number, rather than trying to just roll high: the intuitive understanding that "bonuses are good" breaks down.
In contrast, if you choose to let luck be post-roll and variable (ie, you can choose to add from 0 to your max), then the curve looks like this: (EDIT - yes, I know I'm not being super-precise on these numbers. They're close enough for me, for this post, though.)
Without luck: 4,5,6,7,8,9, 1/2, 1/3, 1/4, 1/5, 1/1/2, or = 64.1%
+1 (variable) luck: 3,4,5,6,7,8,9, 1/2, 1/3, 1/4, 1/5, 1/1/2 = 74.1%
+2 (Variable) luck = 2,3,4,5,6,7,8,9, 1/2, 1/3, 1/4, 1/5, 1/1/2, = 84.1%
+3 (Variable) luck = 1, 2,3,4,5,6,7,8,9 = basically 90%
Note on the last roll, a 1 does count here: 1+9 = 10, which I believe means you have hit a Discovery, even though you need to still re-roll to see what you actually get in ADDITION to the discovery. And to be fair to OneShot, he feels that this curve, at the extreme ends, is simply too easy. I can see his point, but my usual response is "meh. You're taking warp anyway, and this is a pretty optimized character to do exactly this. Let 'em have it."