I don't really think Direct mode is necessarily better.
It uses Timewarp which can help reduce the difference between where your head is when the image is finally presented to the screen and where it was when the frame started to be drawn.
But I find the prediction used when vertical sync is on is really accurate. Though apparently not on some people's computers (*). So maybe they would benefit from Timewarp.
But Timewarp can be used in non-direct mode as well. It's just maybe harder to implement without using Direct Mode. Thing is, Timewarp isn't a true representation of where your head is either. It's also based on prediction, but using a smaller prediction time step and rotating the image to compensate, distorting the image as well though this is probably imperceptible.
(*) By the way, as mentioned before, anyone who found lag with vertical sync enables, please make sure you don't have "force triple buffering" switched on in your drivers. This will definitely cause lag without bringing any benefits.