Partly because they are running tiny cheap power supplies pushed near their limit, and also because many of the actual LEDs they use aren't nearly as efficient as they could be. Also available surface area, in order to fit in most standard bulb locations they need to be small leaving little room for adequately sized heat fins.
yes theres multiple kinds of inefficiencies concerning the external circuitry etc, but there is also the intrinsic inefficiency of current LEDs: some of the high energy photon are downcoverted to a low energy photons by phosphors, so there is non-radiative (heat) generation in the phosphor itself. In theory it is perfectly possible to use "quantum cutting" i.e. 1 photon -> 2 photons instead of 1 photon -> heat + photon. There is quite a lot of research on quantum cutting phosphors, both for LED's and for VUV excitation etc, so these higher efficiencies will probably come, but I don't know how far out in the future. If someone knows of commercially available quantum cutting phosphors (or products using them) I would be very happy to know about them for some physics experiments...
