Just a couple more factors in the comparison.
The 5995, and also the 5645 are based on the standard Hitec digital servo, and are programmable by the HFP-10 for endpoint, deadband, zero, etc…
The 5995 does not have a heatsink, nor does it have any motor cutoff if the motor has high current for too long.
While the 5995 previously represented the best perfomance top end to the Hitec digital range, the 5990 is so much better, it would now be best to avoid the 5995 unless you really want the ability to program with the HFP-10.
The 5990 and 5980 are HMI based servos. They comply to HMI v1.1, which besides normal PWM operation, the can also read position by PWM, change performance parameter sets by PWM, and are serial controllable as well. However they are not programmable by the HFP-10, and Hitec have not yet made the HMI serial programming software available.
The SCC-32 cannot use any of the additional HMI features, but other controllers can, and maybe the SCC-24 will too.
The capability to get feedback from the servos for load sensing is becoming much more commonly used in bipeds, that it is worth having if you are spending a lot of money on servos, even if you don’t use it initially
I have tested the 5990 and it does indeed have overload protection and full HMI v1.1. I beleive the 5980 will also. The heatsink is also an advantage to the 5990, since the aluminium body is more conductive, so you can feel more quickly if a servo is under heavy load and warming up
The 5990 is noisy, and the noise is at rather a high pitch. Some find this annoying.
The only robot that I am aware of using the 5990 is Rooks Pawn 3 from Matt Bauer. Maybe he could give you some input on living with these in a biped.
Finally, you might also consider the Open Servo based Gearhead motor servo that is being developed.