I still don't get it. There are 2 cases:
1. No bias. We accept new blocks that are -30 min or +30 min relative to our time.
2. Bias. We accept new blocks that are -2 hours or +30 min relative to our time.
That's what you mean by bias, right?
But what is the difference between these 2 cases? In both cases miners try to set time as far in the future as they can with all the issues that you and me have just described.
The difference is, in a nutshell, that in the second case I can "go back" up to 2 hours in order to rework my history to form a longer chain that lands within +30min, and in the first case I can only "go back" 30 minutes. Right there it should be easy to objectively say that case 2 is inherently less secure.
Also, for one more specific example, in either case we likely get the "oscillation" effect that DeepCrypto brought up, but in the case #2 there is the incentive to "game" this oscillation where in case #1 the oscillation is always "counterbalanced." (Again this is slightly less of a concern because the distribution is not finite but is still a concern, particularly in the "short term" while we are still in the diminishing distribution phase.)