I was somewhat drunk when I recommended 24/18(2) and it is very obvious to me now that a play with 24/21 8/5(2) is streets ahead. However I would still expect 24/18(2) to be better than 24/21(2) 8/5(2) and so it is, although the narrowness of victory surprises me.
On 3-ply evaluation GNU hates 24/18(2), rating it much worse than 24/21(2) 8/5(2). However on rollout it has it as a better play - only just, but outside the margin of error of 0.05.
The failure to make the 5 point and the additional places to put spares are less important in contrast to 24/21(2) 8/5(2) for two reasons. Firstly, any potential hit opportunities will probably come later, when the midpoint is cleared, so that there will be plenty of time to make the 5 point in the meantime. Secondly, a lot of victories are expected to come from pure racing, where having made the 5 point is irrelevant. One downside of 24/21(2) is that if you roll big numbers and achieve a racing lead, the back men are going to be tough to extract.
Interestingly, 24/18(2) was rated worse on cubeless rollout, but better on cubeful. You can see that it's actually rated better even though it achieves less victories and less gammons (although gammon losses are lower too). No doubt this is because the victories the holding game leads to are crushing ones with inefficient double/drop cube turns, while racing victories after making the bar give the possibility of efficient cubes.