Alright. Hmm, so, we have no empirical data so far.
Indeed - and there was no empirical data presented to "justify" the penalty (or whatever we call it - the *name* of it is not an issue to me - so call it "a rose" but I still think it is a kludge and I am not convinced it will help with the security of the network).
By "fragmenting" the forks all over the place it will actually reduce the effectiveness of such attacks.
Why is that? Malicious nodes could spread forks all over the place and generate blocks very easily on top (because it's PoS not PoW) at the same time to suppress branches from other nodes.
The issue I am referring to is perhaps best illustrated like this:
A1 - B
A2 - B
A1 - C
A2 - D
So assuming that A1 and A2 have the "same weight" then B here is trying to keep both A1 and A2 branches going whilst C and D have each only built on the 1 branch.
Now here is a possible next step:
A1 - B
A2 - B - E
A1 - C
A2 - D - F
Here E and F have separately chosen A2 - B and A2 - D as being "their first choice" and only built upon those (being non-malicious nodes).
So now B's attempt to continue the A1 branch has failed (that fork will likely die now) simply because E did not try to forge 2 blocks at the same height.
Now if we were instead to use the "penalty" approach then in fact we would instead have this result:
A1 - B - E
A2 - B - E
A1 - C - F
A2 - D - F
As both E and F know they are going to get penalised by one or other branch they will forge on both branches causing the branching to continue.
I hope this helps clarify what I am seeing and why I think the idea of "only forging the 1 block at the one height" is a better approach (sure people can "cheat" but as shown provided those cheating are in the minority the effect is greatly reduced).