A few ideas that occurred to me offhand:

-- long hours could be legitimately beneficial to startups, without being beneficial to companies in general

-- willingness to work long hours could be correlated with other traits that make the worker do well for he company, such as enthusiasm

-- related, willingness to work long hours may simply be correlated with other traits that lead the worker to work for a startup at all. For instance, young people without families are more willing to work long hours, and also more willing to take risky jobs such as at starups.

-- long hours could be part of a race towards the bottom, where any individual company that has its workers working longer hours does better, but if all the competing companies do it, they're all worse off than if nobody does it. This could happen if working long hours lets the company get a larger share of a finite resource (such as market share) without increasing the size of that resource

-- signalling competitions between individual workers could lead to the workers all working longer. The absence of middle managers only eliminates certain types of signalling, not all signalling