Well, that's a bit much...
Well yeah unless the wizard who invented the spell was named Daern.
What I meant was that if you could get breakthrough/Insight points simply from any beneficial Experimental result, then there'd be little reason to do Original Research - as experiments are much easier to do, and (if it was just "any beneficial side effect") yield results with less risk than Original Research does. What's the chance - 10% of getting a "Beneficial side effect?", or something like that, per season of Experimentation? That's pretty good, compared to Original Research (serf's parma - didn't someone do the math on the probabilities a while ago?)
Integration would still work: assuming you have a Source, then you know more-or-less what Insight you're going to get; in contrast, Original Research is a tough row to hoe. You get targeted research for it, but that's about it.
In contrast, Experimental Results are like crowdsourcing. Everyone can do it, and as long as folks publish their (random Insights), it's considerably more effective. If it were more of a 1% chance, off on one possible result of the Beneficial Side Effects table though - that seems more reasonable.
But, again - I also agree that Experimental Side Effects would balance better if you can't reproduce their effects via lab notes - but, as someone else mentioned: Ars isn't balanced.
Heh, ya think? 
The only reason Ars is "not balanced" is because it's not tightly defined. And that invites each Troupe to balance it according to their preferences, to what they feel is "fun".
Ars is not a "no-brainer" game like some other RPG's I could name. It's not "turn-key" - open the book and start rolling dice and kiiing monsters. Ars requires some thought, some creative input and thoughtful choices to make it come alive and fulfill its potential. But that "input" is not rigidly defined, and that's the beauty of it. Some players would wholly welcome a "high-fantasy" Major Benefit - like the example above - others (like myself, I freely admit) would not be as comfortable playing in a Mythic Europe like that.
But to each their own - "fun" is as you define it, not as it's defined for you by others. (No, not even by me.) 
So, true - if you are not willing to put some small thought into keeping the game balanced (as you define that), it can easily start to get a little top-heavy, one way or another.
But back OT - I thought of something over an earlier suggestion...
This type of result, which would imply a "regio" within the tower, is uncomfortable to some. ysmv. But "more room" is always welcome, certainly.
Inserting harsh reality into the discussion, stone towers (even "magically created" ones) would have either thick stone arches/ceilings/floors between stories, or notches for heavy beams that would support wooden boards as the ceilings/floors - which are noisy/dusty/etc., but might allow an extra story to be squeezed in. The number of stories you believe you can squeeze in varies - a no-penalty lab needs 10' high ceilings (plus structure between stories, so ~11-12'+), but living quarters can (and almost always historically did) have much lower headroom, no problem.
So, have the tower give them more room without the need for extra-dimensional tangent - the "side benefit" does 2 things:
-
The benefit creates incredibly thin but amply strong stone (marble), that takes up almost no space - not between stories, not for the walls, so "100%" of the listed dimensions are usable (which is a HUGE improvement over the example tower*, with 2-3' thick outer walls making the interior MUCH smaller, etc.) This gives all the privacy, quiet, security and safety of stone without the typical heavy construction required for such.
-
Also, as very thin marble, the walls are translucent, allowing natural light in the day and (optionally) causing the tower to glow from within at night, adding to Aesthetics and being pretty durn impressive all-around.
Toss in the 2 "basement" levels (in the foundations) mentioned before, and you're set.
[i](* The floor area of a 30' (9.1m) diameter tower with 2' (65 cm) stone walls is ~530 square feet (48.7 m[sup]2[/sup]), enough for one standard 500 sq. ft Lab per level of the Tower (which is what the authors have said the tower was designed for!) plus some cramped stair access (which is usually kinda important to include). That's enough for the lab, but not anything else - not the mage's living quarters, not storage, not a servant - not a doghouse, not a murphy bed - nuthin'.
Dropping those walls to (almost) "0" is about 700 square feet (64 m[sup]2[/sup]), or enough for the lab, stair access outside a (wood-thin) wall, and 100-150 square feet (9-14 m[sup]2[/sup]) of room beyond that - a mage's quarters, a small barracks room for multiple grogs, multiple smaller "cell" sized rooms, etc.)[/i]
To me, that is certainly a "Major" side benefit - but some Troupes might prefer something flashier.
I would say that the rules state that the benefit in the spell is transferable.
I would also say that ezzelino is right and that this can cause some really weird stuff to escalate (as long as we assume that magi experiment, and this might not be the case).
To reconcile both, maybe it could be said that side benefits are NOT transferable but that they add to a stabilization of the small spell breakthrough/optimization. That would make publishing them worthwhile for the purposes of research, but would not break the setting with flying towers all around on a pinch if someone got that result by accident.
I also think that the effects of experimentation are tranferable, if one uses the lab text as is.
However, if a magus tries to experiment again starting from an experimental lab text, the benefits will simply not stack unto the results of the new experimentation. Essentially, any new experimentation starts from the base guidelines. The lab text provides some help, but the particulars are lost, since you are deviating from the lab text by experimenting again. So you cannot attain the variations that were obtained on the first version.
That retains the flavour of experimentaiton, while avoiding the problem of escalating power of successive experimentations.