I agree. Standard UPSes convert AC to DC and back, in order to power the server PSUs and these conversions are inefficient. If the battery is in the server (after the PSU), you can miss out these conversions and save a lot of power.

I'd also submit that low-power electronic components are far more readily available (i.e. cheap) than ones which can handle a lot of juice.

Could also be to do with management, particularly switching - switching large DC currents is problematic (i.e. it arcs). It's easier to manage on a smaller scale.