Here's the rest of the story. MacPherson saw this anomaly first with a 10.08mm SWC. Penetration went suddenly trough the gelatin block at 1065 fps, when it should have been approximately a third less. This made him to arrange a special test setup for the configuration. He fired 5.54mm steel SWCs in a sabot, in order to obtain higher velocities without any deformation. The transition area of this phenomenon began at 1065 fps in the first test and with the 5.56mm bullet it began at 1100-1500 fps. Data was erratic in the transition area and one bullet trajectory was sharply curved, others straight.
He explains the theory: "The test results show that cylinders are stable in soft solid penetration at zero angle of attack because the trajectories are straight and the penetration dispersions are low. It logically follows that a symmetric bullet with a flat nose area that is "large enough" will also be stable at zero angle of attack, but may have smaller stability range than a cylinder." [...]
"The drag force on the bullet is minimum only when the bullet is near the perfect alignment condition that has been described as zero angle of attack. The drag force on the bullet during penetration increases significantly as the angle of attack increases." [...]
[...] "The existence of this stable angle of attack [non-zero] was verified by firing soft lead alloy 10.08mm semi-wadcutter bullets at velocities high enough to get some deformation (800 to 1100 ft/sec). The deformation on the shoulder of these semi-wadcutter bullets was obvious and clearly asymmetrical. These bullets had an angle of attack during penetration that was large enough for the bullet nose to shield part of the shoulder from the flow pressure." [...]
"It is not clear why the zero angle of attack point is stable throughout the wound track for the 5.54mm semi-wadcutter configuration when the initial velocity is above 1500 ft/sec. It is also not clear why a non-zero angle of attack is stable for the 5.54mm semi-wadcutter configuration when the initial velocity is below 1100 ft/sec. The combination of these two scenarios is even less explicable. It seems probable that both orientations are stable over small ranges of angle of attack, but consistency of stable point selection depending on initial velocity is unexplained. The erratic penetration in the transition velocity region is not surprising."
"It seems certain that the only stable angle of attack is non-zero for round nose, truncated cone, and 45° cone bullet configurations." [...] "The curved trajectories of these configurations are strong evidence of a stable non-zero angle of attack, but do not quantify the angle of attack contribution to drag. The angle of attack contribution to drag must be large to explain the otherwise mysterious low value of CDC [drag coefficient in the cavitation regime] for spheres (which cannot have any angle of attack effects)." [...]
"Cylinders are stable at zero angle of attack. Bullet configurations with flat nose diameter near the caliber (e.g. Keith semi-wadcutters) are probably stable at zero angle of attack and the values of CDC can be modeled as cylinders with a diameter of the flat nose face." [...]
--
Difficult to estimate if the stability explains this phenomenon completely. Interestingly, 44man also used ultra-hard boolits.