I think you're misunderstanding the problem as this isn't about transaction selection. The issue is, should someone who is mining a more difficult block (say, 2x the normal difficulty) continue mining against an older block even after hearing about a 1x new block? To a naïve first-order approximation, they stand to benefit from doing so because a 2x block would beat out the 1x block and the would be able to steal the fees of the transactions in the 1x block. Of course there are a lot of assumptions wrapped up in that naïve assessment, nor is it clear that it is a reflectively stable outcome (if the 1x miners knew you would do this, how would that change their strategy? how would that change of strategy affect the profitability of this attack? etc.)
jonny1000's comment raises an interesting point.
According to the selected polynomial, we have very different ratios of normalized difficulties between smallest and biggest blocks (around x6 in first model, around x20 for n=100/p=0.99).
Am I wrong if I say that, with the latter, a chain of 6 (or 7, ...) smalls blocks should be considered as less secure, since a single big block could orphan them ?
Usually, people consider a transaction "safe" after 6 blocks. With this model, we really should think in term of work produced after the block embedding the tx.
I guess the choice of the polynomial plays an important role to mitigate this effect.