TL;DR: Minor argument in favor of Haml, followed by rant about intellectually dishonest benchmarking use.
I think adoption rate is a pretty powerful argument in favor of Haml, provided you think your project will ever have collaborators beyond yourself. I have found more design help that has some Haml knowledge than Slim thus far (admittedly in a small sample size).
The Rant on Benchmarking
I wouldn’t worry too much about performance when making this choice. In the benchmark shown there, the percent difference looks large, but the actual amount of time is tiny (a fifth of a second deficit over 1000 template rendering iterations, or if my math is correct, .0002 seconds per template rendering). I have a huge issue with benchmarking like this, because it strikes me as intellectually dishonest to point at these numbers and pretend they actually mean something (again, if my math is way off here, open to being un-convinced). The author actually increased the sample in the second test run from 1,000 iterations to 100,000 iterations, and the performance hit declined on a per template-load basis.
Even if Slim were faster in any meaningful way, if template rendering speed is one of your major performance battles (meaning you’ve defeated asset load time, database queries, etc.) you’ve already won the war.